[ { "name": "bitnet-embedding-0.6b", "company": "microsoft", "release_date": "2026-07-15", "model_release_at": { "value": "2026-07-15", "source": "HF Hub", "source_url": "https://huggingface.co/microsoft/bitnet-embedding-0.6b", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-07-15", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/microsoft/bitnet-embedding-0.6b", "description": "", "category": "Major Release", "slug": "bitnet-embedding-0-6b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "bitnet-embedding-270m", "company": "microsoft", "release_date": "2026-07-15", "model_release_at": { "value": "2026-07-15", "source": "HF Hub", "source_url": "https://huggingface.co/microsoft/bitnet-embedding-270m", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-07-15", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/microsoft/bitnet-embedding-270m", "description": "", "category": "Major Release", "slug": "bitnet-embedding-270m", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Nemotron-3-Embed-1B-NVFP4", "company": "nvidia", "release_date": "2026-07-14", "model_release_at": { "value": "2026-07-14", "source": "HF Hub", "source_url": "https://huggingface.co/nvidia/Nemotron-3-Embed-1B-NVFP4", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-07-14", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/nvidia/Nemotron-3-Embed-1B-NVFP4", "description": "", "category": "Major Release", "slug": "nemotron-3-embed-1b-nvfp4", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "UniVR-34B-Planning", "company": "ByteDance", "release_date": "2026-07-13", "model_release_at": { "value": "2026-07-13", "source": "HF Hub", "source_url": "https://huggingface.co/ByteDance/UniVR-34B-Planning", "retrieved_at": "2026-07-13T12:06:12Z", "verification": "lab-reported", "model_release_at": "2026-07-13", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/ByteDance/UniVR-34B-Planning", "description": "", "category": "Major Release", "slug": "univr-34b-planning", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "nvDock", "company": "nvidia", "release_date": "2026-07-08", "model_release_at": { "value": "2026-07-08", "source": "HF Hub", "source_url": "https://huggingface.co/nvidia/nvDock", "retrieved_at": "2026-07-08T13:03:08Z", "verification": "lab-reported", "model_release_at": "2026-07-08", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/nvidia/nvDock", "description": "", "category": "Major Release", "slug": "nvdock", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "CWIP-1.0", "company": "nvidia", "release_date": "2026-07-07", "model_release_at": { "value": "2026-07-07", "source": "HF Hub", "source_url": "https://huggingface.co/nvidia/CWIP-1.0", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-07-07", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/nvidia/CWIP-1.0", "description": "", "category": "Major Release", "slug": "cwip-1-0", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "HARC-Qwen2.5-7B-Instruct", "company": "microsoft", "release_date": "2026-07-02", "model_release_at": { "value": "2026-07-02", "source": "HF Hub", "source_url": "https://huggingface.co/microsoft/HARC-Qwen2.5-7B-Instruct", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-07-02", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/microsoft/HARC-Qwen2.5-7B-Instruct", "description": "", "category": "Major Release", "slug": "harc-qwen2-5-7b-instruct", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Leanstral-1.5-119B-A6B", "company": "mistralai", "release_date": "2026-07-01", "model_release_at": { "value": "2026-07-01", "source": "HF Hub", "source_url": "https://huggingface.co/mistralai/Leanstral-1.5-119B-A6B", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-07-01", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/mistralai/Leanstral-1.5-119B-A6B", "description": "", "category": "Major Release", "slug": "leanstral-1-5-119b-a6b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "DeepSeek-V4-Flash-DSpark", "company": "deepseek-ai", "release_date": "2026-06-27", "model_release_at": { "value": "2026-06-27", "source": "HF Hub", "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-DSpark", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-27", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-DSpark", "description": "", "category": "Major Release", "slug": "deepseek-v4-flash-dspark", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "DeepSeek-V4-Pro-DSpark", "company": "deepseek-ai", "release_date": "2026-06-27", "model_release_at": { "value": "2026-06-27", "source": "HF Hub", "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-DSpark", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-27", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-DSpark", "description": "", "category": "Major Release", "slug": "deepseek-v4-pro-dspark", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Kimi-K2.6-DFlash", "company": "nvidia", "release_date": "2026-06-23", "model_release_at": { "value": "2026-06-23", "source": "HF Hub", "source_url": "https://huggingface.co/nvidia/Kimi-K2.6-DFlash", "retrieved_at": "2026-07-13T12:06:12Z", "verification": "lab-reported", "model_release_at": "2026-06-23", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/nvidia/Kimi-K2.6-DFlash", "description": "", "category": "Major Release", "slug": "kimi-k2-6-dflash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "EvoQuality", "company": "ByteDance", "release_date": "2026-06-10", "model_release_at": { "value": "2026-06-10", "source": "HF Hub", "source_url": "https://huggingface.co/ByteDance/EvoQuality", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-10", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/ByteDance/EvoQuality", "description": "", "category": "Major Release", "slug": "evoquality", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude Fable 5", "company": "Anthropic", "release_date": "2026-06-09", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "92.6%", "HLE": "53.3%", "SciCode": "60.2%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "62.9%", "IF-Bench": "63.5%", "LiveCodeBench Reasoning": "70.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "description": "Anthropic Claude Fable 5 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "claude-fable-5", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$10.00", "output_per_mtok": "$50.00", "cache_read_per_mtok": "$1.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "92.6%", "HLE": "53.3%", "SciCode": "60.2%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "62.9%", "IF-Bench": "63.5%", "LiveCodeBench Reasoning": "70.0%" }, "links": { "blog": "https://www.anthropic.com/news/claude-fable-5-mythos-5" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "North Mini Code", "company": "Cohere", "release_date": "2026-06-09", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "75.7%", "HLE": "9.9%", "SciCode": "38.2%", "TAU2-bench": "37.4%", "TerminalBench-Hard": "31.1%", "IF-Bench": "57.6%", "LiveCodeBench Reasoning": "32.3%" }, "availability": "Open Source", "announcement_url": "", "description": "Cohere North Mini Code — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "north-mini-code", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "64K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "75.7%", "HLE": "9.9%", "SciCode": "38.2%", "TAU2-bench": "37.4%", "TerminalBench-Hard": "31.1%", "IF-Bench": "57.6%", "LiveCodeBench Reasoning": "32.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "diffusiongemma-26B-A4B-it", "company": "google", "release_date": "2026-06-09", "model_release_at": { "value": "2026-06-09", "source": "HF Hub", "source_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-09", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "description": "", "category": "Major Release", "slug": "diffusiongemma-26b-a4b-it", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-12B-it-qat-q4_0-gguf", "company": "google", "release_date": "2026-06-05", "model_release_at": { "value": "2026-06-05", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-12B-it-qat-q4_0-gguf", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-06-05", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-12B-it-qat-q4_0-gguf", "description": "", "category": "Major Release", "slug": "gemma-4-12b-it-qat-q4-0-gguf", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemma 4 12B", "company": "Google", "release_date": "2026-06-03", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "75.3%", "HLE": "14.6%", "SciCode": "38.2%", "TAU2-bench": "34.8%", "TerminalBench-Hard": "18.2%", "IF-Bench": "73.5%", "LiveCodeBench Reasoning": "55.3%" }, "availability": "Open Source", "announcement_url": "", "description": "Google Gemma 4 12B — 131K tokens context, reasoning model.", "category": "Major Release", "slug": "gemma-4-12b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "131K tokens", "max_output_tokens": "200K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "75.3%", "HLE": "14.8%", "SciCode": "38.2%", "TAU2-bench": "36.3%", "TerminalBench-Hard": "18.2%", "IF-Bench": "73.5%", "LiveCodeBench Reasoning": "55.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "new capabilities to GPT-Rosalind", "company": "OpenAI", "release_date": "2026-06-03", "model_release_at": { "value": "2026-06-03", "source": "Lab announcement (RSS)", "source_url": "https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-03", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind", "description": "", "category": "Major Release", "slug": "new-capabilities-to-gpt-rosalind", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "HARC", "company": "microsoft", "release_date": "2026-06-02", "model_release_at": { "value": "2026-06-02", "source": "HF Hub", "source_url": "https://huggingface.co/microsoft/HARC", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-02", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/microsoft/HARC", "description": "", "category": "Major Release", "slug": "harc", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-E2B-it-qat-mobile-transformers", "company": "google", "release_date": "2026-06-02", "model_release_at": { "value": "2026-06-02", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-E2B-it-qat-mobile-transformers", "retrieved_at": "2026-07-13T12:06:12Z", "verification": "lab-reported", "model_release_at": "2026-06-02", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-E2B-it-qat-mobile-transformers", "description": "", "category": "Major Release", "slug": "gemma-4-e2b-it-qat-mobile-transformers", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-E4B-it-qat-mobile-transformers", "company": "google", "release_date": "2026-06-02", "model_release_at": { "value": "2026-06-02", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-E4B-it-qat-mobile-transformers", "retrieved_at": "2026-07-13T12:06:12Z", "verification": "lab-reported", "model_release_at": "2026-06-02", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-E4B-it-qat-mobile-transformers", "description": "", "category": "Major Release", "slug": "gemma-4-e4b-it-qat-mobile-transformers", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.7 Plus", "company": "Alibaba", "release_date": "2026-06-01", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "90.0%", "HLE": "33.4%", "SciCode": "45.5%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "47.0%", "IF-Bench": "78.0%", "LiveCodeBench Reasoning": "65.0%" }, "availability": "API Only", "announcement_url": "", "description": "Alibaba Qwen3.7 Plus — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-7-plus", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "65K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.40", "output_per_mtok": "$1.16", "cache_read_per_mtok": "$0.08" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "90.0%", "HLE": "33.4%", "SciCode": "45.5%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "47.0%", "IF-Bench": "78.0%", "LiveCodeBench Reasoning": "65.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "MiniMax-M3", "company": "MiniMax", "release_date": "2026-06-01", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "92.9%", "HLE": "37.1%", "SciCode": "45.4%", "TAU2-bench": "88.9%", "TerminalBench-Hard": "42.4%", "IF-Bench": "82.9%", "LiveCodeBench Reasoning": "74.0%" }, "availability": "API Only", "announcement_url": "https://www.minimax.io/models/text/m3", "description": "MiniMax MiniMax-M3 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "minimax-m3", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "511K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.30", "output_per_mtok": "$1.20", "cache_read_per_mtok": "$0.06" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "92.9%", "HLE": "37.1%", "SciCode": "45.4%", "TAU2-bench": "88.9%", "TerminalBench-Hard": "42.4%", "IF-Bench": "82.9%", "LiveCodeBench Reasoning": "74.0%" }, "links": { "blog": "https://www.minimax.io/models/text/m3" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Mellum2", "company": "Hugging Face", "release_date": "2026-06-01", "model_release_at": { "value": "2026-06-01", "source": "Lab announcement (RSS)", "source_url": "https://huggingface.co/blog/JetBrains/mellum2-launch", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-06-01", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/blog/JetBrains/mellum2-launch", "description": "", "category": "Major Release", "slug": "mellum2", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-E2B-it-qat-mobile-ct", "company": "google", "release_date": "2026-06-01", "model_release_at": { "value": "2026-06-01", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-E2B-it-qat-mobile-ct", "retrieved_at": "2026-07-13T12:06:12Z", "verification": "lab-reported", "model_release_at": "2026-06-01", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-E2B-it-qat-mobile-ct", "description": "", "category": "Major Release", "slug": "gemma-4-e2b-it-qat-mobile-ct", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Step 3.7 Flash", "company": "StepFun", "release_date": "2026-05-29", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "80.9%", "HLE": "19.9%", "SciCode": "40.0%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "35.6%", "IF-Bench": "67.3%", "LiveCodeBench Reasoning": "63.7%" }, "availability": "Open Source", "announcement_url": "", "description": "StepFun Step 3.7 Flash — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "step-3-7-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "256K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.20", "output_per_mtok": "$1.15", "cache_read_per_mtok": "$0.04" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "80.9%", "HLE": "19.9%", "SciCode": "40.0%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "35.6%", "IF-Bench": "67.3%", "LiveCodeBench Reasoning": "63.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude Opus 4.8", "company": "Anthropic", "release_date": "2026-05-28", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "92.0%", "HLE": "45.7%", "SciCode": "53.5%", "TAU2-bench": "94.4%", "TerminalBench-Hard": "58.3%", "IF-Bench": "62.2%", "LiveCodeBench Reasoning": "67.7%" }, "availability": "API Only", "announcement_url": "", "description": "Anthropic Claude Opus 4.8 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "claude-opus-4-8", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$6.25", "output_per_mtok": "$25.00", "cache_read_per_mtok": "$0.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "92.0%", "HLE": "45.7%", "SciCode": "53.5%", "TAU2-bench": "94.4%", "TerminalBench-Hard": "58.3%", "IF-Bench": "62.2%", "LiveCodeBench Reasoning": "67.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "LFM2.5-8B-A1B", "company": "Liquid AI", "release_date": "2026-05-28", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "51.3%", "HLE": "6.9%", "SciCode": "7.8%", "TAU2-bench": "16.1%", "TerminalBench-Hard": "4.5%", "IF-Bench": "55.6%", "LiveCodeBench Reasoning": "0.0%" }, "availability": "Open Source", "announcement_url": "", "description": "Liquid AI LFM2.5-8B-A1B — 32K tokens context, reasoning model.", "category": "Major Release", "slug": "lfm2-5-8b-a1b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "32K tokens", "max_output_tokens": "32K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "lfm 1.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "51.3%", "HLE": "6.9%", "SciCode": "7.8%", "TAU2-bench": "16.1%", "TerminalBench-Hard": "4.5%", "IF-Bench": "55.6%", "LiveCodeBench Reasoning": "0.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "HyperNova 60B 2605", "company": "Multiverse Computing", "release_date": "2026-05-26", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "73.3%", "HLE": "15.1%", "SciCode": "33.0%", "TAU2-bench": "63.2%", "TerminalBench-Hard": "23.5%", "IF-Bench": "66.5%", "LiveCodeBench Reasoning": "31.7%" }, "availability": "Open Source", "announcement_url": "", "description": "Multiverse Computing HyperNova 60B 2605 — 131K tokens context, reasoning model.", "category": "Major Release", "slug": "hypernova-60b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "131K tokens", "max_output_tokens": "131K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.04", "output_per_mtok": "$0.14" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "73.3%", "HLE": "15.1%", "SciCode": "33.0%", "TAU2-bench": "63.2%", "TerminalBench-Hard": "23.5%", "IF-Bench": "66.5%", "LiveCodeBench Reasoning": "31.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "MiniCPM5-1B", "company": "OpenBMB", "release_date": "2026-05-25", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "26.9%", "HLE": "4.6%", "SciCode": "1.4%", "TAU2-bench": "82.5%", "TerminalBench-Hard": "0.0%", "IF-Bench": "35.2%", "LiveCodeBench Reasoning": "4.7%" }, "availability": "Open Source", "announcement_url": "", "description": "OpenBMB MiniCPM5-1B — 128K tokens context.", "category": "Major Release", "slug": "minicpm5-1b-non-reasoning", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "128K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "apache-2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "27.8%", "HLE": "6.5%", "SciCode": "4.4%", "TAU2-bench": "81.0%", "TerminalBench-Hard": "0.0%", "IF-Bench": "49.3%", "LiveCodeBench Reasoning": "3.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.7 Max", "company": "Alibaba", "release_date": "2026-05-19", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "92.3%", "HLE": "38.1%", "SciCode": "48.8%", "TAU2-bench": "94.7%", "TerminalBench-Hard": "50.8%", "IF-Bench": "80.5%", "LiveCodeBench Reasoning": "69.0%" }, "availability": "API Only", "announcement_url": "", "description": "Alibaba Qwen3.7 Max — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-7-max", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "65K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$2.50", "output_per_mtok": "$7.50", "cache_read_per_mtok": "$0.25" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "92.3%", "HLE": "38.1%", "SciCode": "48.8%", "TAU2-bench": "94.7%", "TerminalBench-Hard": "50.8%", "IF-Bench": "80.5%", "LiveCodeBench Reasoning": "69.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemini 3.5 Flash", "company": "Google", "release_date": "2026-05-19", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "92.2%", "HLE": "41.0%", "SciCode": "53.1%", "TAU2-bench": "95.3%", "TerminalBench-Hard": "40.9%", "IF-Bench": "76.3%", "LiveCodeBench Reasoning": "69.3%" }, "availability": "API Only", "announcement_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/", "description": "Google Gemini 3.5 Flash — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "gemini-3-5-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$1.50", "output_per_mtok": "$9.00", "cache_read_per_mtok": "$0.15" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "82.8%", "HLE": "23.1%", "SciCode": "48.8%", "TAU2-bench": "58.8%", "TerminalBench-Hard": "46.2%", "IF-Bench": "47.3%", "LiveCodeBench Reasoning": "53.3%" }, "links": { "blog": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "the Ettin Reranker Family", "company": "Hugging Face", "release_date": "2026-05-19", "model_release_at": { "value": "2026-05-19", "source": "Lab announcement (RSS)", "source_url": "https://huggingface.co/blog/ettin-reranker", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-05-19", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/blog/ettin-reranker", "description": "", "category": "Major Release", "slug": "the-ettin-reranker-family", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "SAME-L", "company": "stabilityai", "release_date": "2026-05-17", "model_release_at": { "value": "2026-05-17", "source": "HF Hub", "source_url": "https://huggingface.co/stabilityai/SAME-L", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-05-17", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/stabilityai/SAME-L", "description": "", "category": "Major Release", "slug": "same-l", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "JT-35B-Flash", "company": "China Mobile", "release_date": "2026-05-14", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "82.9%", "HLE": "6.1%", "SciCode": "29.1%", "TAU2-bench": "99.1%", "TerminalBench-Hard": "28.8%", "IF-Bench": "42.0%", "LiveCodeBench Reasoning": "55.3%" }, "availability": "API Only", "announcement_url": "", "description": "China Mobile JT-35B-Flash — 256K tokens context.", "category": "Major Release", "slug": "jt-35b-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "82.9%", "HLE": "6.1%", "SciCode": "29.1%", "TAU2-bench": "99.1%", "TerminalBench-Hard": "28.8%", "IF-Bench": "42.0%", "LiveCodeBench Reasoning": "55.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "MiniCPM-V 4.6 1.3B", "company": "OpenBMB", "release_date": "2026-05-11", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "30.5%", "HLE": "4.9%", "SciCode": "2.1%", "TAU2-bench": "87.7%", "TerminalBench-Hard": "0.0%", "IF-Bench": "26.7%", "LiveCodeBench Reasoning": "6.3%" }, "availability": "Open Source", "announcement_url": "", "description": "OpenBMB MiniCPM-V 4.6 1.3B — 262K tokens context.", "category": "Major Release", "slug": "minicpm-v4-6-1-3b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "262K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "30.5%", "HLE": "4.9%", "SciCode": "2.1%", "TAU2-bench": "87.7%", "TerminalBench-Hard": "0.0%", "IF-Bench": "26.7%", "LiveCodeBench Reasoning": "6.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Ring-2.6-1T", "company": "InclusionAI", "release_date": "2026-05-08", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "85.7%", "HLE": "18.3%", "SciCode": "42.4%", "TAU2-bench": "92.4%", "TerminalBench-Hard": "28.8%", "IF-Bench": "44.6%", "LiveCodeBench Reasoning": "64.3%" }, "availability": "Open Source", "announcement_url": "", "description": "InclusionAI Ring-2.6-1T — 262K tokens context, reasoning model.", "category": "Major Release", "slug": "ring-2-6-1t", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "65K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "MIT", "pricing": { "input_per_mtok": "$0.30", "output_per_mtok": "$2.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "85.7%", "HLE": "18.3%", "SciCode": "42.4%", "TAU2-bench": "92.4%", "TerminalBench-Hard": "28.8%", "IF-Bench": "44.6%", "LiveCodeBench Reasoning": "64.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GPT-5.5 Instant", "company": "OpenAI", "release_date": "2026-05-05", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "84.6%", "HLE": "20.3%", "SciCode": "50.3%", "TAU2-bench": "49.4%", "TerminalBench-Hard": "42.4%", "IF-Bench": "71.5%", "LiveCodeBench Reasoning": "55.7%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/gpt-5-5-instant", "description": "OpenAI GPT-5.5 Instant — 400K tokens context, reasoning model.", "category": "Major Release", "slug": "gpt-5-5-instant-05-26", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "400K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-08-31", "license": "Proprietary", "pricing": { "input_per_mtok": "$5.00", "output_per_mtok": "$30.00", "cache_read_per_mtok": "$0.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "84.6%", "HLE": "20.3%", "SciCode": "50.3%", "TAU2-bench": "49.4%", "TerminalBench-Hard": "42.4%", "IF-Bench": "71.5%", "LiveCodeBench Reasoning": "55.7%" }, "links": { "blog": "https://openai.com/index/gpt-5-5-instant" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-31B-it-qat-q4_0-gguf", "company": "google", "release_date": "2026-05-01", "model_release_at": { "value": "2026-05-01", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-31B-it-qat-q4_0-gguf", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-05-01", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-31B-it-qat-q4_0-gguf", "description": "", "category": "Major Release", "slug": "gemma-4-31b-it-qat-q4-0-gguf", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "gemma-4-26B-A4B-it-qat-q4_0-gguf", "company": "google", "release_date": "2026-05-01", "model_release_at": { "value": "2026-05-01", "source": "HF Hub", "source_url": "https://huggingface.co/google/gemma-4-26B-A4B-it-qat-q4_0-gguf", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-05-01", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/google/gemma-4-26B-A4B-it-qat-q4_0-gguf", "description": "", "category": "Major Release", "slug": "gemma-4-26b-a4b-it-qat-q4-0-gguf", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Grok 4.3", "company": "xAI", "release_date": "2026-04-30", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "90.1%", "HLE": "35.0%", "SciCode": "47.3%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "37.9%", "IF-Bench": "81.3%", "LiveCodeBench Reasoning": "64.3%" }, "availability": "API Only", "announcement_url": "https://docs.x.ai/developers/models/grok-4.3", "description": "xAI Grok 4.3 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "grok-4-3", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "131K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$1.25", "output_per_mtok": "$2.50", "cache_read_per_mtok": "$0.20" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "90.1%", "HLE": "35.0%", "SciCode": "47.3%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "37.9%", "IF-Bench": "81.3%", "LiveCodeBench Reasoning": "64.3%" }, "links": { "blog": "https://docs.x.ai/developers/models/grok-4.3" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Granite 4.1 30B", "company": "IBM", "release_date": "2026-04-29", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "48.1%", "HLE": "4.2%", "SciCode": "25.8%", "TAU2-bench": "42.1%", "TerminalBench-Hard": "2.3%", "IF-Bench": "44.4%", "LiveCodeBench Reasoning": "18.7%" }, "availability": "Open Source", "announcement_url": "", "description": "IBM Granite 4.1 30B — 131K tokens context.", "category": "Major Release", "slug": "granite-4-1-30b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "131K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "48.1%", "HLE": "4.2%", "SciCode": "25.8%", "TAU2-bench": "42.1%", "TerminalBench-Hard": "2.3%", "IF-Bench": "44.4%", "LiveCodeBench Reasoning": "18.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Granite 4.1 3B", "company": "IBM", "release_date": "2026-04-29", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "31.4%", "HLE": "3.4%", "SciCode": "11.9%", "TAU2-bench": "19.6%", "TerminalBench-Hard": "2.3%", "IF-Bench": "33.7%", "LiveCodeBench Reasoning": "3.0%" }, "availability": "Open Source", "announcement_url": "", "description": "IBM Granite 4.1 3B — 131K tokens context.", "category": "Major Release", "slug": "granite-4-1-3b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "131K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "31.4%", "HLE": "3.4%", "SciCode": "11.9%", "TAU2-bench": "19.6%", "TerminalBench-Hard": "2.3%", "IF-Bench": "33.7%", "LiveCodeBench Reasoning": "3.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Granite 4.1 8B", "company": "IBM", "release_date": "2026-04-29", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "43.3%", "HLE": "3.8%", "SciCode": "21.8%", "TAU2-bench": "27.8%", "TerminalBench-Hard": "0.0%", "IF-Bench": "38.6%", "LiveCodeBench Reasoning": "12.0%" }, "availability": "Open Source", "announcement_url": "", "description": "IBM Granite 4.1 8B — 131K tokens context.", "category": "Major Release", "slug": "granite-4-1-8b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "131K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.05", "output_per_mtok": "$0.10" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "43.3%", "HLE": "3.8%", "SciCode": "21.8%", "TAU2-bench": "27.8%", "TerminalBench-Hard": "0.0%", "IF-Bench": "38.6%", "LiveCodeBench Reasoning": "12.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Mistral Medium 3.5", "company": "Mistral", "release_date": "2026-04-29", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "74.8%", "HLE": "12.8%", "SciCode": "39.6%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "33.3%", "IF-Bench": "68.8%", "LiveCodeBench Reasoning": "61.0%" }, "availability": "Open Source", "announcement_url": "", "description": "Mistral Mistral Medium 3.5 — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "mistral-medium-3-5", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Other", "pricing": { "input_per_mtok": "$1.50", "output_per_mtok": "$7.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "74.8%", "HLE": "12.8%", "SciCode": "39.6%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "33.3%", "IF-Bench": "68.8%", "LiveCodeBench Reasoning": "61.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "DeepSeek V4 Flash", "company": "DeepSeek", "release_date": "2026-04-24", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "89.4%", "HLE": "32.1%", "SciCode": "44.9%", "TAU2-bench": "95.0%", "TerminalBench-Hard": "35.6%", "IF-Bench": "79.2%", "LiveCodeBench Reasoning": "63.0%" }, "availability": "Open Source", "announcement_url": "https://api-docs.deepseek.com/news/news260424", "description": "DeepSeek DeepSeek V4 Flash — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "deepseek-v4-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "64K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "MIT", "pricing": { "input_per_mtok": "$0.14", "output_per_mtok": "$0.28", "cache_read_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "89.4%", "HLE": "32.1%", "SciCode": "44.9%", "TAU2-bench": "95.0%", "TerminalBench-Hard": "35.6%", "IF-Bench": "79.2%", "LiveCodeBench Reasoning": "63.0%" }, "links": { "blog": "https://api-docs.deepseek.com/news/news260424" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "DeepSeek V4 Pro", "company": "DeepSeek", "release_date": "2026-04-24", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "88.8%", "HLE": "35.9%", "SciCode": "50.0%", "TAU2-bench": "96.2%", "TerminalBench-Hard": "46.2%", "IF-Bench": "76.5%", "LiveCodeBench Reasoning": "66.3%" }, "availability": "Open Source", "announcement_url": "https://api-docs.deepseek.com/news/news260424", "description": "DeepSeek DeepSeek V4 Pro — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "deepseek-v4-pro", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "393K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "MIT", "pricing": { "input_per_mtok": "$1.74", "output_per_mtok": "$3.48", "cache_read_per_mtok": "$0.01" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "88.8%", "HLE": "35.9%", "SciCode": "50.0%", "TAU2-bench": "96.2%", "TerminalBench-Hard": "46.2%", "IF-Bench": "76.5%", "LiveCodeBench Reasoning": "66.3%" }, "links": { "blog": "https://api-docs.deepseek.com/news/news260424" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Ling-2.6-1T", "company": "InclusionAI", "release_date": "2026-04-23", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "75.2%", "HLE": "8.2%", "SciCode": "37.0%", "TAU2-bench": "89.8%", "TerminalBench-Hard": "31.1%", "IF-Bench": "56.9%", "LiveCodeBench Reasoning": "34.7%" }, "availability": "Open Source", "announcement_url": "", "description": "InclusionAI Ling-2.6-1T — 262K tokens context.", "category": "Major Release", "slug": "ling-2-6-1t", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "32K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Mit", "pricing": { "input_per_mtok": "$0.30", "output_per_mtok": "$2.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "75.2%", "HLE": "8.2%", "SciCode": "37.0%", "TAU2-bench": "89.8%", "TerminalBench-Hard": "31.1%", "IF-Bench": "56.9%", "LiveCodeBench Reasoning": "34.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GPT-5.5", "company": "OpenAI", "release_date": "2026-04-23", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "93.5%", "HLE": "44.3%", "SciCode": "56.1%", "TAU2-bench": "93.9%", "TerminalBench-Hard": "60.6%", "IF-Bench": "75.9%", "LiveCodeBench Reasoning": "74.3%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/introducing-gpt-5-5/", "description": "OpenAI GPT-5.5 — 922K tokens context, reasoning model.", "category": "Major Release", "slug": "gpt-5-5", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "922K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$5.00", "output_per_mtok": "$30.00", "cache_read_per_mtok": "$0.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "93.5%", "HLE": "44.3%", "SciCode": "56.1%", "TAU2-bench": "93.9%", "TerminalBench-Hard": "60.6%", "IF-Bench": "75.9%", "LiveCodeBench Reasoning": "74.3%" }, "links": { "blog": "https://openai.com/index/introducing-gpt-5-5/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Hy3-preview", "company": "Tencent", "release_date": "2026-04-23", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "86.7%", "HLE": "25.5%", "SciCode": "41.2%", "TAU2-bench": "92.7%", "TerminalBench-Hard": "34.1%", "IF-Bench": "63.1%", "LiveCodeBench Reasoning": "54.7%" }, "availability": "Open Source", "announcement_url": "", "description": "Tencent Hy3-preview — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "hy3", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "256K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "TENCENT HY COMMUNITY LICENSE AGREEMENT", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00", "cache_read_per_mtok": "$0.10" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "86.7%", "HLE": "25.5%", "SciCode": "41.2%", "TAU2-bench": "92.7%", "TerminalBench-Hard": "34.1%", "IF-Bench": "63.1%", "LiveCodeBench Reasoning": "54.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "CADD-Base-7B", "company": "apple", "release_date": "2026-04-23", "model_release_at": { "value": "2026-04-23", "source": "HF Hub", "source_url": "https://huggingface.co/apple/CADD-Base-7B", "retrieved_at": "2026-07-08T12:10:39Z", "verification": "lab-reported", "model_release_at": "2026-04-23", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/apple/CADD-Base-7B", "description": "", "category": "Major Release", "slug": "cadd-base-7b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.6 27B", "company": "Alibaba", "release_date": "2026-04-22", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "84.2%", "HLE": "21.6%", "SciCode": "39.8%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "34.8%", "IF-Bench": "67.6%", "LiveCodeBench Reasoning": "68.7%" }, "availability": "Open Source", "announcement_url": "", "description": "Alibaba Qwen3.6 27B — 262K tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-6-27b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "81K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.60", "output_per_mtok": "$3.60" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "84.2%", "HLE": "21.6%", "SciCode": "39.8%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "34.8%", "IF-Bench": "67.6%", "LiveCodeBench Reasoning": "68.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "MiMo-V2.5", "company": "Xiaomi", "release_date": "2026-04-22", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "84.9%", "HLE": "25.2%", "SciCode": "43.1%", "TAU2-bench": "90.6%", "TerminalBench-Hard": "41.7%", "IF-Bench": "67.1%", "LiveCodeBench Reasoning": "62.7%" }, "availability": "Open Source", "announcement_url": "https://mimo.xiaomi.com/mimo-v2-5/", "description": "Xiaomi MiMo-V2.5 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "mimo-v2-5-0424", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "131K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Mit", "pricing": { "input_per_mtok": "$0.36", "output_per_mtok": "$1.80", "cache_read_per_mtok": "$0.20" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "84.9%", "HLE": "25.2%", "SciCode": "43.1%", "TAU2-bench": "90.6%", "TerminalBench-Hard": "41.7%", "IF-Bench": "67.1%", "LiveCodeBench Reasoning": "62.7%" }, "links": { "blog": "https://mimo.xiaomi.com/mimo-v2-5/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "MiMo-V2.5-Pro", "company": "Xiaomi", "release_date": "2026-04-22", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "86.6%", "HLE": "33.8%", "SciCode": "50.2%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "43.2%", "IF-Bench": "79.9%", "LiveCodeBench Reasoning": "73.3%" }, "availability": "Open Source", "announcement_url": "https://mimo.xiaomi.com/mimo-v2-5-pro/", "description": "Xiaomi MiMo-V2.5-Pro — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "mimo-v2-5-pro", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "131K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Mit", "pricing": { "input_per_mtok": "$1.00", "output_per_mtok": "$3.00", "cache_read_per_mtok": "$0.80" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "86.6%", "HLE": "33.8%", "SciCode": "50.2%", "TAU2-bench": "94.2%", "TerminalBench-Hard": "43.2%", "IF-Bench": "79.9%", "LiveCodeBench Reasoning": "73.3%" }, "links": { "blog": "https://mimo.xiaomi.com/mimo-v2-5-pro/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Ling 2.6 Flash", "company": "InclusionAI", "release_date": "2026-04-21", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "59.3%", "HLE": "6.2%", "SciCode": "27.1%", "TAU2-bench": "86.0%", "TerminalBench-Hard": "21.2%", "IF-Bench": "57.4%", "LiveCodeBench Reasoning": "25.0%" }, "availability": "Open Source", "announcement_url": "", "description": "InclusionAI Ling 2.6 Flash — 262K tokens context.", "category": "Major Release", "slug": "ling-2-6-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "32K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Mit", "pricing": { "input_per_mtok": "$0.10", "output_per_mtok": "$0.30", "cache_read_per_mtok": "$0.02" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "59.3%", "HLE": "6.2%", "SciCode": "27.1%", "TAU2-bench": "86.0%", "TerminalBench-Hard": "21.2%", "IF-Bench": "57.4%", "LiveCodeBench Reasoning": "25.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.6 Max Preview", "company": "Alibaba", "release_date": "2026-04-20", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "88.8%", "HLE": "28.9%", "SciCode": "46.9%", "TAU2-bench": "95.9%", "TerminalBench-Hard": "43.9%", "IF-Bench": "76.6%", "LiveCodeBench Reasoning": "69.7%" }, "availability": "API Only", "announcement_url": "https://qwen.ai/blog?id=qwen3.6-max-preview", "description": "Alibaba Qwen3.6 Max Preview — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-6-max", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "65K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$1.30", "output_per_mtok": "$7.80", "cache_read_per_mtok": "$0.13" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "88.8%", "HLE": "28.9%", "SciCode": "46.9%", "TAU2-bench": "95.9%", "TerminalBench-Hard": "43.9%", "IF-Bench": "76.6%", "LiveCodeBench Reasoning": "69.7%" }, "links": { "blog": "https://qwen.ai/blog?id=qwen3.6-max-preview" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Kimi K2.6", "company": "Kimi", "release_date": "2026-04-20", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "91.1%", "HLE": "35.9%", "SciCode": "53.5%", "TAU2-bench": "95.9%", "TerminalBench-Hard": "43.9%", "IF-Bench": "76.0%", "LiveCodeBench Reasoning": "69.7%" }, "availability": "Open Source", "announcement_url": "https://www.kimi.com/blog/kimi-k2-6", "description": "Kimi Kimi K2.6 — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "kimi-k2-6", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "256K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Modified MIT", "pricing": { "input_per_mtok": "$0.95", "output_per_mtok": "$4.00", "cache_read_per_mtok": "$0.16" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "91.1%", "HLE": "35.9%", "SciCode": "53.5%", "TAU2-bench": "95.9%", "TerminalBench-Hard": "43.9%", "IF-Bench": "76.0%", "LiveCodeBench Reasoning": "69.7%" }, "links": { "blog": "https://www.kimi.com/blog/kimi-k2-6" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.6 35B A3B", "company": "Alibaba", "release_date": "2026-04-16", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "84.1%", "HLE": "20.2%", "SciCode": "35.8%", "TAU2-bench": "95.3%", "TerminalBench-Hard": "34.8%", "IF-Bench": "64.4%", "LiveCodeBench Reasoning": "63.7%" }, "availability": "Open Source", "announcement_url": "", "description": "Alibaba Qwen3.6 35B A3B — 262K tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-6-35b-a3b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "65K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.25", "output_per_mtok": "$1.49" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "84.1%", "HLE": "20.2%", "SciCode": "35.8%", "TAU2-bench": "95.3%", "TerminalBench-Hard": "34.8%", "IF-Bench": "64.4%", "LiveCodeBench Reasoning": "63.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude Opus 4.7", "company": "Anthropic", "release_date": "2026-04-16", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "91.4%", "HLE": "39.6%", "SciCode": "54.5%", "TAU2-bench": "88.6%", "TerminalBench-Hard": "51.5%", "IF-Bench": "58.6%", "LiveCodeBench Reasoning": "70.3%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-opus-4-7", "description": "Anthropic Claude Opus 4.7 — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "claude-opus-4-7", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2026-01-01", "license": "Proprietary", "pricing": { "input_per_mtok": "$6.25", "output_per_mtok": "$25.00", "cache_read_per_mtok": "$0.50" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "91.4%", "HLE": "39.6%", "SciCode": "54.5%", "TAU2-bench": "88.6%", "TerminalBench-Hard": "51.5%", "IF-Bench": "58.6%", "LiveCodeBench Reasoning": "70.3%" }, "links": { "blog": "https://www.anthropic.com/news/claude-opus-4-7" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "JT-MINI", "company": "China Mobile", "release_date": "2026-04-15", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "67.6%", "HLE": "6.6%", "SciCode": "27.2%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "18.2%", "IF-Bench": "36.7%", "LiveCodeBench Reasoning": "11.7%" }, "availability": "API Only", "announcement_url": "", "description": "China Mobile JT-MINI — 128K tokens context.", "category": "Major Release", "slug": "jt-mini", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "67.6%", "HLE": "6.6%", "SciCode": "27.2%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "18.2%", "IF-Bench": "36.7%", "LiveCodeBench Reasoning": "11.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "nemotron-labs-audio-visual-flamingo-hf", "company": "nvidia", "release_date": "2026-04-15", "model_release_at": { "value": "2026-04-15", "source": "HF Hub", "source_url": "https://huggingface.co/nvidia/nemotron-labs-audio-visual-flamingo-hf", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-04-15", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/nvidia/nemotron-labs-audio-visual-flamingo-hf", "description": "", "category": "Major Release", "slug": "nemotron-labs-audio-visual-flamingo-hf", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "skala-1.1", "company": "microsoft", "release_date": "2026-04-13", "model_release_at": { "value": "2026-04-13", "source": "HF Hub", "source_url": "https://huggingface.co/microsoft/skala-1.1", "retrieved_at": "2026-07-20T11:45:18Z", "verification": "lab-reported", "model_release_at": "2026-04-13", "benchmark_released_at": null }, "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": {}, "availability": "API", "announcement_url": "https://huggingface.co/microsoft/skala-1.1", "description": "", "category": "Major Release", "slug": "skala-1-1", "tagline": null, "architecture": null, "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [], "output_modalities": [], "knowledge_cutoff": null, "license": null, "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": {}, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "EXAONE 4.5 33B", "company": "LG AI Research", "release_date": "2026-04-09", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "79.4%", "HLE": "11.6%", "SciCode": "28.0%", "TAU2-bench": "78.1%", "TerminalBench-Hard": "20.5%", "IF-Bench": "58.0%", "LiveCodeBench Reasoning": "49.3%" }, "availability": "Open Source", "announcement_url": "", "description": "LG AI Research EXAONE 4.5 33B — 262K tokens context, reasoning model.", "category": "Major Release", "slug": "exaone-4-5-33b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "262K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "EXAONE AI Model License Agreement 1.2 - NC", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "79.4%", "HLE": "11.6%", "SciCode": "28.0%", "TAU2-bench": "78.1%", "TerminalBench-Hard": "20.5%", "IF-Bench": "58.0%", "LiveCodeBench Reasoning": "49.3%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Muse Spark", "company": "Meta", "release_date": "2026-04-08", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "88.4%", "HLE": "39.9%", "SciCode": "51.5%", "TAU2-bench": "91.5%", "TerminalBench-Hard": "45.5%", "IF-Bench": "75.9%", "LiveCodeBench Reasoning": "69.7%" }, "availability": "API Only", "announcement_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/", "description": "Meta Muse Spark — 262K tokens context, reasoning model.", "category": "Major Release", "slug": "muse-spark", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "262K tokens", "max_output_tokens": "131K", "input_modalities": [ "text", "image", "audio" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "88.4%", "HLE": "39.9%", "SciCode": "51.5%", "TAU2-bench": "91.5%", "TerminalBench-Hard": "45.5%", "IF-Bench": "75.9%", "LiveCodeBench Reasoning": "69.7%" }, "links": { "blog": "https://ai.meta.com/blog/introducing-muse-spark-msl/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GLM-5.1", "company": "Z AI", "release_date": "2026-04-07", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "86.8%", "HLE": "28.0%", "SciCode": "43.8%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "43.2%", "IF-Bench": "76.3%", "LiveCodeBench Reasoning": "62.3%" }, "availability": "Open Source", "announcement_url": "https://huggingface.co/zai-org/GLM-5.1", "description": "Z AI GLM-5.1 — 200K tokens context, reasoning model.", "category": "Major Release", "slug": "glm-5-1", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "131K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Mit", "pricing": { "input_per_mtok": "$1.40", "output_per_mtok": "$4.40", "cache_read_per_mtok": "$0.26" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "86.8%", "HLE": "28.0%", "SciCode": "43.8%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "43.2%", "IF-Bench": "76.3%", "LiveCodeBench Reasoning": "62.3%" }, "links": { "blog": "https://huggingface.co/zai-org/GLM-5.1" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Grok 4.20 0309 v2", "company": "xAI", "release_date": "2026-04-07", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "91.1%", "HLE": "32.2%", "SciCode": "45.6%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "37.9%", "IF-Bench": "81.2%", "LiveCodeBench Reasoning": "58.0%" }, "availability": "API Only", "announcement_url": "", "description": "xAI Grok 4.20 0309 v2 — 2M tokens context, reasoning model.", "category": "Major Release", "slug": "grok-4-20", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "2M tokens", "max_output_tokens": "131K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$2.00", "output_per_mtok": "$6.00", "cache_read_per_mtok": "$1.10" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "91.1%", "HLE": "32.2%", "SciCode": "45.6%", "TAU2-bench": "93.0%", "TerminalBench-Hard": "37.9%", "IF-Bench": "81.2%", "LiveCodeBench Reasoning": "58.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Solar Pro 3", "company": "Upstage", "release_date": "2026-04-06", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "72.4%", "HLE": "10.1%", "SciCode": "24.7%", "TAU2-bench": "86.3%", "TerminalBench-Hard": "7.6%", "IF-Bench": "71.2%", "LiveCodeBench Reasoning": "27.0%" }, "availability": "API Only", "announcement_url": "", "description": "Upstage Solar Pro 3 — 128K tokens context, reasoning model.", "category": "Major Release", "slug": "solar-pro-3", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "131K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "72.4%", "HLE": "10.1%", "SciCode": "24.7%", "TAU2-bench": "86.3%", "TerminalBench-Hard": "7.6%", "IF-Bench": "71.2%", "LiveCodeBench Reasoning": "27.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemma 4 E4B", "company": "Google", "release_date": "2026-04-03", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "57.6%", "HLE": "3.7%", "SciCode": "24.4%", "TAU2-bench": "20.8%", "TerminalBench-Hard": "8.3%", "IF-Bench": "44.2%", "LiveCodeBench Reasoning": "30.7%" }, "availability": "Open Source", "announcement_url": "", "description": "Google Gemma 4 E4B — 128K tokens context, reasoning model.", "category": "Major Release", "slug": "gemma-4-e4b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.30", "output_per_mtok": "$1.25" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "57.6%", "HLE": "3.7%", "SciCode": "24.4%", "TAU2-bench": "20.8%", "TerminalBench-Hard": "8.3%", "IF-Bench": "44.2%", "LiveCodeBench Reasoning": "30.7%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.6 Plus", "company": "Alibaba", "release_date": "2026-04-02", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "88.2%", "HLE": "25.7%", "SciCode": "40.7%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "43.9%", "IF-Bench": "75.2%", "LiveCodeBench Reasoning": "69.7%" }, "availability": "API Only", "announcement_url": "https://qwen.ai/blog?id=qwen3.6", "description": "Alibaba Qwen3.6 Plus — 1M tokens context, reasoning model.", "category": "Major Release", "slug": "qwen3-6-plus", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "65K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.50", "output_per_mtok": "$3.00", "cache_read_per_mtok": "$0.05" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "88.2%", "HLE": "25.7%", "SciCode": "40.7%", "TAU2-bench": "97.7%", "TerminalBench-Hard": "43.9%", "IF-Bench": "75.2%", "LiveCodeBench Reasoning": "69.7%" }, "links": { "blog": "https://qwen.ai/blog?id=qwen3.6" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemma 4 26B A4B", "company": "Google", "release_date": "2026-04-02", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "79.2%", "HLE": "18.3%", "SciCode": "40.0%", "TAU2-bench": "43.6%", "TerminalBench-Hard": "13.6%", "IF-Bench": "72.4%", "LiveCodeBench Reasoning": "55.7%" }, "availability": "Open Source", "announcement_url": "https://aistudio.google.com/app/prompts/new_chat?model=gemma-4-26b-a4b-it", "description": "Google Gemma 4 26B A4B — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "gemma-4-26b-a4b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.13", "output_per_mtok": "$0.40", "cache_read_per_mtok": "$0.10" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "79.2%", "HLE": "18.3%", "SciCode": "40.0%", "TAU2-bench": "43.6%", "TerminalBench-Hard": "13.6%", "IF-Bench": "72.4%", "LiveCodeBench Reasoning": "55.7%" }, "links": { "blog": "https://aistudio.google.com/app/prompts/new_chat?model=gemma-4-26b-a4b-it" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemma 4 31B", "company": "Google", "release_date": "2026-04-02", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "85.7%", "HLE": "22.7%", "SciCode": "43.4%", "TAU2-bench": "59.9%", "TerminalBench-Hard": "36.4%", "IF-Bench": "75.6%", "LiveCodeBench Reasoning": "62.0%" }, "availability": "Open Source", "announcement_url": "https://aistudio.google.com/app/prompts/new_chat?model=gemma-4-31b-it", "description": "Google Gemma 4 31B — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "gemma-4-31b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "85.7%", "HLE": "22.7%", "SciCode": "43.4%", "TAU2-bench": "59.9%", "TerminalBench-Hard": "36.4%", "IF-Bench": "75.6%", "LiveCodeBench Reasoning": "62.0%" }, "links": { "blog": "https://aistudio.google.com/app/prompts/new_chat?model=gemma-4-31b-it" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemma 4 E2B", "company": "Google", "release_date": "2026-04-02", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "43.3%", "HLE": "4.8%", "SciCode": "20.9%", "TAU2-bench": "20.8%", "TerminalBench-Hard": "3.0%", "IF-Bench": "38.0%", "LiveCodeBench Reasoning": "15.0%" }, "availability": "Open Source", "announcement_url": "", "description": "Google Gemma 4 E2B — 128K tokens context, reasoning model.", "category": "Major Release", "slug": "gemma-4-e2b", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "128K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "43.3%", "HLE": "4.8%", "SciCode": "20.9%", "TAU2-bench": "20.8%", "TerminalBench-Hard": "3.0%", "IF-Bench": "38.0%", "LiveCodeBench Reasoning": "15.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Step 3.5 Flash 2603", "company": "StepFun", "release_date": "2026-04-02", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "82.6%", "HLE": "22.6%", "SciCode": "38.5%", "TAU2-bench": "87.4%", "TerminalBench-Hard": "32.6%", "IF-Bench": "66.5%", "LiveCodeBench Reasoning": "54.3%" }, "availability": "API Only", "announcement_url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash", "description": "StepFun Step 3.5 Flash 2603 — 256K tokens context, reasoning model.", "category": "Major Release", "slug": "step-3-5-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "256K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00", "cache_read_per_mtok": "$0.02" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "82.6%", "HLE": "22.6%", "SciCode": "38.5%", "TAU2-bench": "87.4%", "TerminalBench-Hard": "32.6%", "IF-Bench": "66.5%", "LiveCodeBench Reasoning": "54.3%" }, "links": { "blog": "https://huggingface.co/stepfun-ai/Step-3.5-Flash" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Trinity Large Thinking", "company": "Arcee AI", "release_date": "2026-04-01", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "75.2%", "HLE": "14.7%", "SciCode": "36.1%", "TAU2-bench": "90.1%", "TerminalBench-Hard": "22.7%", "IF-Bench": "56.3%", "LiveCodeBench Reasoning": "33.0%" }, "availability": "Open Source", "announcement_url": "https://www.arcee.ai/blog/trinity-large-thinking", "description": "Arcee AI Trinity Large Thinking — 512K tokens context, reasoning model.", "category": "Major Release", "slug": "trinity-large-thinking", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "512K tokens", "max_output_tokens": "80K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Apache 2.0", "pricing": { "input_per_mtok": "$0.23", "output_per_mtok": "$0.88", "cache_read_per_mtok": "$0.15" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "75.2%", "HLE": "14.7%", "SciCode": "36.1%", "TAU2-bench": "90.1%", "TerminalBench-Hard": "22.7%", "IF-Bench": "56.3%", "LiveCodeBench Reasoning": "33.0%" }, "links": { "blog": "https://www.arcee.ai/blog/trinity-large-thinking" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GLM 5V Turbo", "company": "Z AI", "release_date": "2026-04-01", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "80.9%", "HLE": "15.8%", "SciCode": "43.5%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "32.6%", "IF-Bench": "61.1%", "LiveCodeBench Reasoning": "61.0%" }, "availability": "API Only", "announcement_url": "https://docs.z.ai/guides/vlm/glm-5v-turbo", "description": "Z AI GLM 5V Turbo — 200K tokens context, reasoning model.", "category": "Major Release", "slug": "glm-5v-turbo", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "131K", "input_modalities": [ "text", "image", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "80.9%", "HLE": "15.8%", "SciCode": "43.5%", "TAU2-bench": "98.5%", "TerminalBench-Hard": "32.6%", "IF-Bench": "61.1%", "LiveCodeBench Reasoning": "61.0%" }, "links": { "blog": "https://docs.z.ai/guides/vlm/glm-5v-turbo" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Qwen3.5 Omni Flash", "company": "Alibaba", "release_date": "2026-03-30", "model_type": "LLM", "parameters": null, "key_features": [], "performance_metrics": { "GPQA Diamond": "74.2%", "HLE": "7.1%", "SciCode": "25.5%", "TAU2-bench": "84.5%", "TerminalBench-Hard": "8.3%", "IF-Bench": "38.0%", "LiveCodeBench Reasoning": "44.0%" }, "availability": "API Only", "announcement_url": "", "description": "Alibaba Qwen3.5 Omni Flash — 256K tokens context.", "category": "Major Release", "slug": "qwen3-5-omni-flash", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "65K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input_per_mtok": "$0.10", "output_per_mtok": "$0.80" }, "providers": [], "rate_limits": null, "benchmarks": { "GPQA Diamond": "74.2%", "HLE": "7.1%", "SciCode": "25.5%", "TAU2-bench": "84.5%", "TerminalBench-Hard": "8.3%", "IF-Bench": "38.0%", "LiveCodeBench Reasoning": "44.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemini 3.1 Pro", "company": "Google", "release_date": "2026-02-19", "model_type": "LLM", "parameters": "~1T MoE", "key_features": [ "2x reasoning improvement", "ARC-AGI-2 score of 77.1%", "Enhanced multimodal understanding", "Deep Think mode" ], "performance_metrics": { "ARC-AGI-2": "77.1%", "MMLU": "93.8%", "MATH": "89.4%" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/ai/", "description": "Google's latest flagship model with a major 2x jump in reasoning capabilities", "category": "Major Release", "slug": "google-gemini-3-1-pro", "tagline": "Google's flagship reasoning model with a 2x jump on hard multi-step tasks.", "architecture": "Sparse Mixture-of-Experts (MoE)", "training_tokens": null, "context_window": "2M tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "audio", "video", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-12", "license": "Proprietary", "pricing": { "input_per_mtok": "$2.50", "output_per_mtok": "$10.00", "cached_input_per_mtok": "$0.25" }, "providers": [ "Google AI Studio", "Vertex AI", "Gemini API" ], "rate_limits": null, "benchmarks": { "ARC-AGI-2": "77.1%", "MMLU-Pro": "93.8%", "MATH": "89.4%", "GPQA Diamond": "84.2%", "SWE-bench Verified": "72.3%", "LiveCodeBench": "78.9%" }, "links": { "blog": "https://blog.google/technology/ai/", "docs": "https://ai.google.dev/gemini-api/docs/models", "model_card": "https://ai.google.dev/gemini-api/docs/models/gemini-3" }, "predecessor_slug": "google-gemini-3-pro", "changelog": [ "2x reasoning score on ARC-AGI-2 vs Gemini 3 Pro", "Context window expanded to 2M tokens", "Deep Think mode enabled by default on the Pro tier", "Lower latency on first-token despite larger context" ], "launch_coverage": [ { "title": "Gemini 3.1 Pro Is A Powerhouse For Deep Work — Here Are 7 Prompts That...", "url": "https://www.tomsguide.com/ai/gemini-3-1-pro-is-a-powerhouse-for-deep-work-here-are-7-prompts-that-prove-it", "source": "tomsguide.com" } ] }, { "name": "Claude Sonnet 4.6", "company": "Anthropic", "release_date": "2026-02-17", "model_type": "LLM", "parameters": "~500B", "key_features": [ "Agent Teams: orchestrate 2-16 Claude instances", "Near-Opus performance at 1/5th cost", "80.8% SWE-bench Verified", "Fast mode research preview" ], "performance_metrics": { "SWE-bench": "80.8%", "MMLU": "92.1%", "HumanEval": "95.2%", "SWE-bench Verified": "80.8%", "GPQA Diamond": "79.9%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news", "description": "Anthropic's latest Sonnet with Agent Teams capability and near-Opus performance at a fraction of the cost", "category": "Major Release", "slug": "anthropic-claude-sonnet-4-6", "tagline": "Near-Opus quality at a fraction of the cost, with Agent Teams orchestration.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "500K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-10", "license": "Proprietary", "pricing": { "input_per_mtok": "$3.00", "output_per_mtok": "$15.00", "cached_input_per_mtok": "$0.30", "cache_read_per_mtok": "$0.30" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "80.8%", "MMLU": "92.1%", "HumanEval": "95.2%", "GPQA Diamond": "79.7%", "AIME 2025": "88.5%", "TAU-bench": "71.2%", "HLE": "10.8%", "SciCode": "44.1%", "TAU2-bench": "78.9%", "TerminalBench-Hard": "42.4%", "IF-Bench": "42.4%", "LiveCodeBench Reasoning": "58.7%" }, "links": { "blog": "https://www.anthropic.com/news", "docs": "https://docs.anthropic.com/en/docs/about-claude/models", "system_card": "https://www.anthropic.com/news/claude-4-family" }, "predecessor_slug": "anthropic-claude-sonnet-4", "changelog": [ "Agent Teams: orchestrate 2–16 Claude instances in parallel", "+8.5pt on SWE-bench Verified vs Sonnet 4", "1/5 the cost of Opus 4.5 at ~95% of coding quality", "Fast mode research preview for lower-latency inference" ], "launch_coverage": [ { "title": "I Ran 7 Real-World Prompts On Gemini 3 And Claude Sonnet 4.6 — The Results...", "url": "https://www.tomsguide.com/ai/i-ran-7-real-world-prompts-on-gemini-3-and-claude-sonnet-4-6-the-results-surprised-me", "source": "tomsguide.com" } ] }, { "name": "DeepSeek V3.2", "company": "DeepSeek", "release_date": "2026-02-12", "model_type": "LLM", "parameters": "671B MoE", "key_features": [ "1M+ token context window (10x expansion)", "Improved reasoning capabilities", "Open source release", "Cost-effective inference" ], "performance_metrics": { "MMLU": "90.1%", "HumanEval": "92.5%", "Context Window": "1M+ tokens", "MMLU-Pro": "83.7%", "GPQA Diamond": "75.1%" }, "availability": "Open Source", "announcement_url": "https://github.com/deepseek-ai/DeepSeek-V3", "description": "Major update with 10x context window expansion to over 1 million tokens", "category": "Update", "slug": "deepseek-deepseek-v3-2", "tagline": "Open-weight MoE with a 1M+ token context window and strong coding.", "architecture": "Sparse MoE (37B active / 671B total)", "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "8K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-09", "license": "DeepSeek License (open weights, commercial OK)", "pricing": { "input_per_mtok": "$0.27", "output_per_mtok": "$1.10", "cached_input_per_mtok": "$0.07", "cache_read_per_mtok": "$0.50" }, "providers": [ "DeepSeek API", "Hugging Face", "Together AI", "Fireworks AI" ], "rate_limits": null, "benchmarks": { "MMLU": "90.1%", "HumanEval": "92.5%", "MATH": "85.6%", "GPQA": "68.4%", "LiveCodeBench": "86.2%", "MMLU-Pro": "86.2%", "GPQA Diamond": "84.0%", "HLE": "22.2%", "SciCode": "38.9%", "TAU2-bench": "90.6%", "TerminalBench-Hard": "35.6%", "IF-Bench": "60.7%", "LiveCodeBench Reasoning": "65.0%" }, "links": { "repo": "https://github.com/deepseek-ai/DeepSeek-V3", "paper": "https://arxiv.org/abs/2412.19437", "model_card": "https://huggingface.co/deepseek-ai/DeepSeek-V3" }, "predecessor_slug": "deepseek-deepseek-v3", "changelog": [ "10x context window expansion (128K → 1M+ tokens)", "Sliding-window attention for long-context throughput", "Improved chain-of-thought reasoning", "Native FP8 inference support" ], "launch_coverage": [] }, { "name": "GLM-5", "company": "Zhipu AI", "release_date": "2026-02-11", "model_type": "LLM", "parameters": "744B", "key_features": [ "First frontier model trained on Huawei Ascend chips (no NVIDIA)", "#1 HLE score (50.4%)", "1.2% hallucination rate via Slime RL", "136x cheaper than Claude Opus 4.5" ], "performance_metrics": { "HLE": "50.4%", "Hallucination Rate": "1.2%", "Cost": "$0.11/M tokens" }, "availability": "API Only", "announcement_url": "https://www.zhipuai.cn/", "description": "First frontier AI model trained entirely without NVIDIA GPUs, using Huawei Ascend chips", "category": "Major Release", "slug": "zhipu-ai-glm-5", "tagline": "First frontier model trained entirely on Huawei Ascend silicon.", "architecture": "Dense Transformer (744B)", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-11", "license": "Proprietary (open weights for non-frontier sizes)", "pricing": { "input_per_mtok": "$0.11", "output_per_mtok": "$0.28" }, "providers": [ "Zhipu BigModel API" ], "rate_limits": null, "benchmarks": { "HLE": "50.4%", "MMLU": "88.7%", "Hallucination Rate": "1.2%", "C-Eval": "92.1%", "GSM8K": "94.8%" }, "links": { "blog": "https://www.zhipuai.cn/", "paper": "https://arxiv.org/abs/2406.12793" }, "predecessor_slug": null, "changelog": [ "Trained entirely on Huawei Ascend 910B clusters (no NVIDIA)", "Slime RL fine-tuning drops hallucination rate to 1.2%", "136x cheaper than Claude Opus 4.5 at comparable quality" ], "launch_coverage": [ { "title": "Glm 5 Is Being Tested On Openrouter", "url": "https://www.reddit.com/r/LocalLLaMA/comments/1qxqpdz/glm_5_is_being_tested_on_openrouter/", "source": "Reddit" } ] }, { "name": "GPT-5.3 Codex", "company": "OpenAI", "release_date": "2026-02-05", "model_type": "Code", "parameters": "~200B", "key_features": [ "Self-improving agentic coding", "25% faster than GPT-5.2-Codex", "1,000+ tokens/sec generation", "First OpenAI model flagged 'high' on cybersecurity framework" ], "performance_metrics": { "Terminal-Bench": "77.3%", "SWE-Bench Pro": "SOTA", "Speed": "1,000+ tok/s", "SWE-bench Verified": "82.4%", "HumanEval": "96.8%", "GPQA Diamond": "91.5%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/", "description": "OpenAI's specialized self-improving coding model with state-of-the-art software engineering performance", "category": "Major Release", "slug": "openai-gpt-5-3-codex", "tagline": "Coding-specialized variant of GPT-5.3, tuned for agentic IDE workflows.", "architecture": "MoE (coding-specialized fine-tune)", "training_tokens": null, "context_window": "400K tokens", "max_output_tokens": "100K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-11", "license": "Proprietary", "pricing": { "input_per_mtok": "$1.25", "output_per_mtok": "$10.00", "cached_input_per_mtok": "$0.13", "cache_read_per_mtok": "$0.17" }, "providers": [ "OpenAI API", "Azure OpenAI", "GitHub Copilot" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "82.4%", "HumanEval": "96.8%", "LiveCodeBench": "84.2%", "Aider Polyglot": "79.5%", "GPQA Diamond": "91.5%", "HLE": "39.9%", "SciCode": "53.2%", "TAU2-bench": "86.0%", "TerminalBench-Hard": "53.0%", "IF-Bench": "75.4%", "LiveCodeBench Reasoning": "74.0%" }, "links": { "blog": "https://openai.com/blog/" }, "predecessor_slug": "openai-gpt-5-2-codex", "changelog": [ "+4pt on SWE-bench Verified vs GPT-5.2 Codex", "Native IDE tool-calling at reduced latency", "Extended max output to 100K for multi-file patches" ], "launch_coverage": [ { "title": "Openai Just Launched Gpt-5.3-Codex: A Faster Agentic Coding Model Unifying Frontier Code Performance And Professional...", "url": "https://www.marktechpost.com/2026/02/05/openai-just-launched-gpt-5-3-codex-a-faster-agentic-coding-model-unifying-frontier-code-performance-and-professional-reasoning-into-one-system/", "source": "MarkTechPost" }, { "title": "Gpt-5.3-Codex System Card", "url": "https://openai.com/index/gpt-5-3-codex-system-card", "source": "OpenAI" }, { "title": "Introducing Gpt-5.3-Codex", "url": "https://openai.com/index/introducing-gpt-5-3-codex", "source": "OpenAI" }, { "title": "They Actually Dropped Gpt-5.3 Codex The Minute Opus 4.6 Dropped Lol", "url": "https://www.reddit.com/r/OpenAI/comments/1qwsnp9/they_actually_dropped_gpt53_codex_the_minute_opus/", "source": "Reddit" }, { "title": "Gpt-5.3 Codex Vs Opus 4.6: We Benchmarked Both On Our Production Rails Codebase — The...", "url": "https://www.reddit.com/r/ClaudeAI/comments/1qxr7vs/gpt53_codex_vs_opus_46_we_benchmarked_both_on_our/", "source": "Reddit" } ] }, { "name": "Kimi K2", "company": "Moonshot AI", "release_date": "2026-01-20", "model_type": "LLM", "parameters": "1.04T MoE", "key_features": [ "First open-weight model #1 on LMSYS Chatbot Arena", "1.04 trillion parameters", "K2.5 agent swarms with up to 100 sub-agents", "$0.15/M input tokens" ], "performance_metrics": { "LMSYS Arena": "#1", "Parameters": "1.04T", "Cost": "$0.15/M tokens" }, "availability": "Open Source", "announcement_url": "https://www.moonshot.cn/", "description": "First open-weight model to rank #1 on LMSYS Chatbot Arena with over 1 trillion parameters", "category": "Major Release", "slug": "moonshot-ai-kimi-k2", "tagline": "Moonshot's open-weight frontier MoE with strong agentic benchmarks.", "architecture": "MoE (32B active / ~1T total)", "training_tokens": null, "context_window": "2M tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-10", "license": "Modified MIT (open weights)", "pricing": { "input_per_mtok": "$0.15", "output_per_mtok": "$2.50" }, "providers": [ "Moonshot API", "Hugging Face", "Together AI" ], "rate_limits": null, "benchmarks": { "MMLU": "91.3%", "SWE-bench Verified": "65.8%", "GPQA Diamond": "74.1%", "LiveCodeBench": "68.9%" }, "links": { "repo": "https://github.com/MoonshotAI/Kimi-K2", "model_card": "https://huggingface.co/moonshotai/Kimi-K2" }, "predecessor_slug": "moonshot-ai-moonshot-kimi", "changelog": [ "2M token context window (20x vs first Kimi)", "Agentic tool-use tuning via MuonClip optimizer", "Open weights under modified MIT" ], "launch_coverage": [ { "title": "Gpt-5.2 Xhigh, Glm-4.7, Kimi K2 Thinking, Deepseek V3.2 On Fresh Swe-Rebench (December 2025)", "url": "https://www.reddit.com/r/LocalLLaMA/comments/1qefa7q/gpt52_xhigh_glm47_kimi_k2_thinking_deepseek_v32/", "source": "Reddit" }, { "title": "Kimi K2 Artificial Analysis Score", "url": "https://www.reddit.com/r/LocalLLaMA/comments/1qos25i/kimi_k2_artificial_analysis_score/", "source": "Reddit" } ] }, { "name": "GPT-5.2 Codex", "company": "OpenAI", "release_date": "2025-12-18", "model_type": "Code", "parameters": "~200B", "key_features": [ "Specialized for software engineering", "Enhanced agentic coding", "Multi-file refactoring", "Advanced debugging capabilities" ], "performance_metrics": { "SWE-Bench": "SOTA", "HumanEval": "96.1%", "Terminal-Bench": "72.8%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/", "description": "Specialized coding variant of GPT-5.2 focused on software engineering tasks", "category": "Update", "slug": "openai-gpt-5-2-codex", "tagline": "Prior-gen Codex variant of GPT-5.2 for agentic coding.", "architecture": "MoE (coding fine-tune)", "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-08", "license": "Proprietary", "pricing": { "input_per_mtok": "$1.50", "output_per_mtok": "$12.00" }, "providers": [ "OpenAI API", "Azure OpenAI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "78.2%", "HumanEval": "95.1%", "LiveCodeBench": "80.4%" }, "links": {}, "predecessor_slug": "openai-gpt-5-1", "changelog": [], "launch_coverage": [ { "title": "Introducing Gpt-5.2-Codex", "url": "https://openai.com/index/introducing-gpt-5-2-codex", "source": "OpenAI" }, { "title": "Addendum To Gpt-5.2 System Card: Gpt-5.2-Codex", "url": "https://openai.com/index/gpt-5-2-codex-system-card", "source": "OpenAI" } ] }, { "name": "Mistral Large 3", "company": "Mistral", "release_date": "2025-12-15", "model_type": "LLM", "parameters": "~123B", "key_features": [ "128K context window", "Improved multilingual capabilities", "Enhanced function calling", "Competitive with GPT-5 class models" ], "performance_metrics": { "MMLU": "88.2%", "HumanEval": "82.5%", "MATH": "72.1%", "MMLU-Pro": "80.7%", "GPQA Diamond": "68.0%" }, "availability": "API Only", "announcement_url": "https://mistral.ai/news/", "description": "Mistral's flagship model competing with GPT-5 class models at a fraction of the cost", "category": "Major Release", "slug": "mistral-mistral-large-3", "tagline": "Mistral's flagship proprietary model, tuned for European enterprise.", "architecture": "Dense Transformer", "training_tokens": null, "context_window": "256K tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-07", "license": "Mistral Commercial License", "pricing": { "input_per_mtok": "$2.00", "output_per_mtok": "$6.00" }, "providers": [ "Mistral La Plateforme", "Azure", "AWS Bedrock" ], "rate_limits": null, "benchmarks": { "MMLU": "89.4%", "HumanEval": "91.2%", "MATH": "82.1%", "MMMU": "76.8%", "MMLU-Pro": "80.7%", "GPQA Diamond": "68.0%", "LiveCodeBench": "46.5%", "HLE": "4.1%", "SciCode": "36.2%", "TAU2-bench": "24.6%", "TerminalBench-Hard": "15.9%", "IF-Bench": "36.2%", "LiveCodeBench Reasoning": "34.7%" }, "links": { "blog": "https://mistral.ai/news/" }, "predecessor_slug": "mistral-mistral-large", "changelog": [], "launch_coverage": [] }, { "name": "GPT-5.2", "company": "OpenAI", "release_date": "2025-12-11", "model_type": "LLM", "parameters": "~200B", "key_features": [ "Enhanced reasoning capabilities", "Improved adaptive reasoning", "Better multimodal understanding", "Faster inference" ], "performance_metrics": { "MMLU": "92.8%", "MATH": "88.5%", "HumanEval": "95.8%", "MMLU-Pro": "87.4%", "SWE-bench Verified": "72.5%", "GPQA Diamond": "90.3%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/", "description": "Iterative improvement on GPT-5.1 with enhanced reasoning and faster performance", "category": "Update", "slug": "openai-gpt-5-2", "tagline": "Late-2025 GPT-5 refresh with improved reasoning and steerability.", "architecture": "MoE", "training_tokens": null, "context_window": "400K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "audio" ], "output_modalities": [ "text", "audio" ], "knowledge_cutoff": "2025-08", "license": "Proprietary", "pricing": { "input_per_mtok": "$2.00", "output_per_mtok": "$10.00", "cache_read_per_mtok": "$0.17" }, "providers": [ "OpenAI API", "Azure OpenAI", "ChatGPT" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "87.4%", "GPQA Diamond": "90.3%", "SWE-bench Verified": "72.5%", "AIME 2025": "92.1%", "LiveCodeBench": "88.9%", "HLE": "35.4%", "SciCode": "52.1%", "TAU2-bench": "84.8%", "TerminalBench-Hard": "47.0%", "IF-Bench": "75.4%", "LiveCodeBench Reasoning": "72.7%" }, "links": {}, "predecessor_slug": "openai-gpt-5-1", "changelog": [], "launch_coverage": [ { "title": "Openai’S Gpt-5.2 ‘Code Red’ Response To Google Is Coming Next Week", "url": "https://www.theverge.com/report/838857/openai-gpt-5-2-release-date-code-red-google-response", "source": "The Verge" }, { "title": "Openai Introduces Gpt 5.2: A Long Context Workhorse For Agents, Coding And Knowledge Work", "url": "https://www.marktechpost.com/2025/12/11/openai-introduces-gpt-5-2-a-long-context-workhorse-for-agents-coding-and-knowledge-work/", "source": "MarkTechPost" }, { "title": "Openai Calls Gpt-5.2 The Best Model Yet For Professionals", "url": "https://www.theverge.com/ai-artificial-intelligence/842529/openai-gpt-5-2-new-model-chatgpt", "source": "The Verge" }, { "title": "Openai Launches Gpt-5.2 As It Navigates ‘Code Red’", "url": "https://www.wired.com/story/openai-gpt-launch-gemini-code-red/", "source": "Wired" }, { "title": "Openai Fires Back At Google With Gpt-5.2 After ‘Code Red’ Memo | Techcrunch", "url": "https://techcrunch.com/2025/12/11/openai-fires-back-at-google-with-gpt-5-2-after-code-red-memo/", "source": "TechCrunch" } ] }, { "name": "Claude Opus 4.5", "company": "Anthropic", "release_date": "2025-11-24", "model_type": "LLM", "parameters": "~500B", "key_features": [ "First model to break 80.9% on SWE-Bench Verified", "67% price reduction vs previous Opus", "Extended reasoning capabilities", "Advanced coding performance" ], "performance_metrics": { "SWE-bench": "80.9%", "MMLU": "91.5%", "HumanEval": "95.0%", "MMLU-Pro": "88.9%", "SWE-bench Verified": "78.9%", "GPQA Diamond": "81.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news", "description": "Anthropic's most capable model with breakthrough coding performance and major price reduction", "category": "Major Release", "slug": "anthropic-claude-opus-4-5", "tagline": "Anthropic's top-tier reasoning model for complex research and agents.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "500K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-08", "license": "Proprietary", "pricing": { "input_per_mtok": "$15.00", "output_per_mtok": "$75.00", "cached_input_per_mtok": "$1.50", "cache_read_per_mtok": "$0.50" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "78.9%", "MMLU": "92.8%", "GPQA Diamond": "86.6%", "AIME 2025": "90.5%", "MMLU-Pro": "89.5%", "LiveCodeBench": "87.1%", "HLE": "28.4%", "SciCode": "49.5%", "TAU2-bench": "89.5%", "TerminalBench-Hard": "47.0%", "IF-Bench": "58.0%", "LiveCodeBench Reasoning": "74.0%" }, "links": { "blog": "https://www.anthropic.com/news" }, "predecessor_slug": "anthropic-claude-opus-4-1", "changelog": [], "launch_coverage": [ { "title": "Epoch Ai Appears To Have Leaked Claude Opus 4.5'S Release Date Is Tomorrow", "url": "https://www.reddit.com/r/ClaudeAI/comments/1p4s2yo/epoch_ai_appears_to_have_leaked_claude_opus_45s/", "source": "Reddit" }, { "title": "I Created A Free Retirement Planner With Claude Opus 4.5", "url": "https://www.reddit.com/r/ClaudeAI/comments/1q1co0q/i_created_a_free_retirement_planner_with_claude/", "source": "Reddit" } ] }, { "name": "Gemini 3 Pro", "company": "Google", "release_date": "2025-11-18", "model_type": "Multimodal", "parameters": "~1T MoE", "key_features": [ "1M token context window", "Deep Think reasoning mode", "Solved 5/6 IMO 2025 problems", "#1 on LMSYS Arena" ], "performance_metrics": { "ARC-AGI": "87.5%", "MMLU": "93.2%", "LMSYS Arena": "#1" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/ai/", "description": "Google's flagship model with Deep Think mode, ranked #1 on LMSYS Arena at launch", "category": "Major Release", "slug": "google-gemini-3-pro", "tagline": "First Gemini 3 tier release; strong multimodal + long-context.", "architecture": "Sparse MoE", "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image", "audio", "video", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-09", "license": "Proprietary", "pricing": { "input_per_mtok": "$2.50", "output_per_mtok": "$10.00" }, "providers": [ "Google AI Studio", "Vertex AI", "Gemini API" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "89.4%", "MMMU": "82.1%", "GPQA Diamond": "78.5%", "SWE-bench Verified": "68.2%" }, "links": { "blog": "https://blog.google/technology/ai/" }, "predecessor_slug": "google-gemini-2-5-flash", "changelog": [], "launch_coverage": [ { "title": "Google To Release Nano Banana Pro Powered By Gemini 3 Pro Next Week - Testingcatalog", "url": "https://www.testingcatalog.com/google-to-release-nano-banana-pro-powered-by-gemini-3-pro-next-week/", "source": "testingcatalog.com" }, { "title": "Google Changes Gemini 3 Pro Free Access Limits Due To ‘High Demand’ - 9To5Google", "url": "http://9to5google.com/2025/11/27/gemini-3-pro-free-limits/", "source": "9to5google.com" }, { "title": "Google Caps Gemini 3 Pro And Nano Banana Pro Usage As Servers Feel The Heat...", "url": "https://www.androidcentral.com/apps-software/insane-gemini-3-pro-demand-forces-google-to-cap-access", "source": "androidcentral.com" }, { "title": "Google Just Dropped A New Agentic Benchmark: Gemini 3 Pro Beat Pokémon Crystal (Defeating Red)...", "url": "https://www.reddit.com/r/singularity/comments/1pngym8/google_just_dropped_a_new_agentic_benchmark/", "source": "Reddit" }, { "title": "[D] How Did Gemini 3 Pro Manage To Get 38.3% On Humanity'S Last Exam?", "url": "https://www.reddit.com/r/MachineLearning/comments/1pgqbjd/d_how_did_gemini_3_pro_manage_to_get_383_on/", "source": "Reddit" } ] }, { "name": "GPT-5.1", "company": "OpenAI", "release_date": "2025-11-12", "model_type": "LLM", "parameters": "~200B", "key_features": [ "Adaptive reasoning modes", "Perfect 100% on AIME 2025", "87.5% on ARC-AGI", "Enhanced multimodal capabilities" ], "performance_metrics": { "ARC-AGI": "87.5%", "AIME 2025": "100%", "MMLU": "92.5%", "MMLU-Pro": "87.0%", "SWE-bench Verified": "70.1%", "GPQA Diamond": "87.3%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/", "description": "Major GPT-5 iteration with adaptive reasoning and perfect scores on math competitions", "category": "Major Release", "slug": "openai-gpt-5-1", "tagline": "Maintenance update to GPT-5 with steerability + latency improvements.", "architecture": "MoE", "training_tokens": null, "context_window": "400K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "audio" ], "output_modalities": [ "text", "audio" ], "knowledge_cutoff": "2025-06", "license": "Proprietary", "pricing": { "input_per_mtok": "$2.25", "output_per_mtok": "$11.00", "cache_read_per_mtok": "$0.12" }, "providers": [ "OpenAI API", "Azure OpenAI", "ChatGPT" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "87.0%", "GPQA Diamond": "87.3%", "SWE-bench Verified": "70.1%", "LiveCodeBench": "86.8%", "HLE": "26.5%", "SciCode": "43.3%", "TAU2-bench": "81.9%", "TerminalBench-Hard": "45.5%", "IF-Bench": "72.9%", "LiveCodeBench Reasoning": "75.0%" }, "links": {}, "predecessor_slug": "openai-gpt-5", "changelog": [], "launch_coverage": [ { "title": "Openai Walks A Tricky Tightrope With Gpt-5.1’S Eight New Personalities", "url": "https://arstechnica.com/ai/2025/11/openai-walks-a-tricky-tightrope-with-gpt-5-1s-eight-new-personalities/", "source": "Ars Technica" }, { "title": "Openai Says The Brand-New Gpt-5.1 Is ‘Warmer’ And Has More ‘Personality’ Options", "url": "https://www.theverge.com/news/802653/openai-gpt-5-1-upgrade-personality-presets", "source": "The Verge" }, { "title": "Openai Introduces Gpt-5.1: Combining Adaptive Reasoning, Account Level Personalization, And Updated Safety Metrics In The...", "url": "https://www.marktechpost.com/2025/11/12/openai-introduces-gpt-5-1-combining-adaptive-reasoning-account-level-personalization-and-updated-safety-metrics-in-the-gpt-5-stack/", "source": "MarkTechPost" }, { "title": "Openai Debuts Gpt-5.1-Codex-Max, A Long-Horizon Agentic Coding Model With Compaction For Multi-Window Workflows", "url": "https://www.marktechpost.com/2025/11/19/openai-debuts-gpt-5-1-codex-max-a-long-horizon-agentic-coding-model-with-compaction-for-multi-window-workflows/", "source": "MarkTechPost" }, { "title": "Gpt-5.1 Instant And Gpt-5.1 Thinking System Card Addendum", "url": "https://openai.com/index/gpt-5-system-card-addendum-gpt-5-1", "source": "OpenAI" } ] }, { "name": "GPT-5", "company": "OpenAI", "release_date": "2025-08-15", "model_type": "LLM", "parameters": "~200B", "key_features": [ "Adaptive reasoning (routes between quick and deep thinking)", "Improved math and coding", "Enhanced multimodal reasoning", "New safety architecture" ], "performance_metrics": { "MMLU": "91.0%", "HumanEval": "93.5%", "MATH": "85.0%", "MMLU-Pro": "87.1%", "SWE-bench Verified": "67.4%", "GPQA Diamond": "85.4%" }, "availability": "API Only", "announcement_url": "https://openai.com/index/", "description": "OpenAI's next-generation flagship model with adaptive reasoning capabilities", "category": "Major Release", "slug": "openai-gpt-5", "tagline": "OpenAI's flagship unified reasoning + chat model replacing the GPT-4 line.", "architecture": "MoE with unified reasoning router", "training_tokens": null, "context_window": "400K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "audio" ], "output_modalities": [ "text", "audio" ], "knowledge_cutoff": "2025-05", "license": "Proprietary", "pricing": { "input_per_mtok": "$2.50", "output_per_mtok": "$12.00", "cache_read_per_mtok": "$0.12" }, "providers": [ "OpenAI API", "Azure OpenAI", "ChatGPT" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "80.6%", "GPQA Diamond": "67.3%", "SWE-bench Verified": "67.4%", "MATH": "90.1%", "HumanEval": "95.1%", "LiveCodeBench": "55.8%", "MATH-500": "86.1%", "AIME 2025": "36.7%", "HLE": "5.4%", "SciCode": "38.8%", "TAU2-bench": "67.0%", "TerminalBench-Hard": "18.2%", "IF-Bench": "45.6%", "LiveCodeBench Reasoning": "25.0%" }, "links": { "blog": "https://openai.com/blog/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [ { "title": "Openai Teases Gpt-5 Launch Event This Thursday", "url": "https://www.theverge.com/news/720114/openai-gpt-5-launch-event-tease", "source": "The Verge" }, { "title": "Openai Launches Gpt-5 Free To All Chatgpt Users", "url": "https://arstechnica.com/ai/2025/08/openai-launches-gpt-5-free-to-all-chatgpt-users/", "source": "Ars Technica" }, { "title": "Openai Just Released Gpt-5: The Smartest, Fastest, And Most Useful Openai Model", "url": "https://www.marktechpost.com/2025/08/07/openai-just-released-gpt-5-the-smartest-fastest-and-most-useful-openai-model/", "source": "MarkTechPost" }, { "title": "Openai Launches Gpt-5, Nano, Mini And Pro — Not Agi, But Capable Of Generating ‘Software-On-Demand’", "url": "https://venturebeat.com/ai/openai-launches-gpt-5-not-agi-but-capable-of-generating-software-on-demand/", "source": "VentureBeat" }, { "title": "The Gpt-5 Feature Openai Hasn’T Talked About (But It Changes Everything) 🧠", "url": "https://www.reddit.com/r/AI_Agents/comments/1mkfcy3/the_gpt5_feature_openai_hasnt_talked_about_but_it/", "source": "Reddit" } ] }, { "name": "Claude Opus 4.1", "company": "Anthropic", "release_date": "2025-07-15", "model_type": "LLM", "parameters": "~500B", "key_features": [ "Improved multi-file refactoring", "Enhanced agentic capabilities", "Better long-context performance", "Reduced hallucinations" ], "performance_metrics": { "SWE-bench": "75.2%", "MMLU": "90.8%", "HumanEval": "94.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news", "description": "Iterative improvement on Claude Opus 4 with enhanced multi-file refactoring", "category": "Update", "slug": "anthropic-claude-opus-4-1", "tagline": "Mid-2025 Opus refresh focused on agentic coding reliability.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "32K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-03", "license": "Proprietary", "pricing": { "input_per_mtok": "$15.00", "output_per_mtok": "$75.00" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "74.5%", "MMLU": "91.2%", "GPQA Diamond": "79.1%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [ { "title": "Claude Opus 4.1 - Gets The Job Done No Matter What The Obstacle.", "url": "https://www.reddit.com/r/ClaudeAI/comments/1mjaxgt/claude_opus_41_gets_the_job_done_no_matter_what/", "source": "Reddit" }, { "title": "Anyone Else Playing \"Bug Whack-A-Mole\" With Claude Opus 4.1? 😅", "url": "https://www.reddit.com/r/ClaudeAI/comments/1mtb2ka/anyone_else_playing_bug_whackamole_with_claude/", "source": "Reddit" } ] }, { "name": "Gemini 2.5 Flash", "company": "Google", "release_date": "2025-06-20", "model_type": "Multimodal", "parameters": "~175B", "key_features": [ "Enhanced image editing stabilization", "Faster inference", "Improved multimodal understanding", "Cost-effective deployment" ], "performance_metrics": { "MMLU": "87.5%", "Speed": "2x Gemini 2.0 Flash", "Image Quality": "High", "MMLU-Pro": "80.9%", "HumanEval": "95.1%", "GPQA Diamond": "68.3%" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/ai/", "description": "Google's fast and cost-effective model with enhanced image capabilities", "category": "Update", "slug": "google-gemini-2-5-flash", "tagline": "Google's cost-optimized multimodal model with thinking mode.", "architecture": "Dense multimodal transformer", "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "65K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-01", "license": "Proprietary", "pricing": { "input_per_mtok": "$0.30", "output_per_mtok": "$2.50", "cache_read_per_mtok": "$0.03" }, "providers": [ "Google AI Studio", "Vertex AI", "Gemini API" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "83.2%", "MMMU": "79.7%", "LiveCodeBench": "69.5%", "GPQA Diamond": "79.0%", "HumanEval": "96.2%", "MATH-500": "98.1%", "AIME 2025": "82.3%", "HLE": "11.1%", "SciCode": "39.4%", "TAU2-bench": "31.6%", "TerminalBench-Hard": "13.6%", "IF-Bench": "50.3%", "LiveCodeBench Reasoning": "61.7%" }, "links": { "blog": "https://blog.google/technology/google-deepmind/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude Sonnet 4", "company": "Anthropic", "release_date": "2025-05-22", "model_type": "LLM", "parameters": "~500B", "key_features": [ "Enhanced reasoning capabilities", "Improved safety measures", "Advanced multimodal understanding", "Extended context window" ], "performance_metrics": { "MMLU": "91.2%", "HumanEval": "94.5%", "MATH": "76.8%" }, "availability": "API Only", "announcement_url": "https://docs.anthropic.com/en/release-notes/claude-apps", "description": "Latest generation Claude model with significant performance improvements", "category": "Major Release", "slug": "anthropic-claude-sonnet-4", "tagline": "Claude 4 mid-tier with strong coding and long-horizon agentic reliability.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2025-03", "license": "Proprietary", "pricing": { "input_per_mtok": "$3.00", "output_per_mtok": "$15.00" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "72.3%", "MMLU": "88.7%", "GPQA Diamond": "74.0%" }, "links": { "blog": "https://www.anthropic.com/news/claude-4", "system_card": "https://www-cdn.anthropic.com/claude-4-system-card.pdf" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude Sonnet 3.7", "company": "Anthropic", "release_date": "2025-02-24", "model_type": "LLM", "parameters": "~300B", "key_features": [ "Improved reasoning", "Better code generation", "Enhanced safety", "Reduced hallucinations" ], "performance_metrics": { "MMLU": "89.5%", "HumanEval": "93.2%", "MATH": "74.1%" }, "availability": "API Only", "announcement_url": "https://docs.anthropic.com/en/release-notes/claude-apps", "description": "Iterative improvement on Claude 3.5 with enhanced capabilities", "category": "Update", "slug": "anthropic-claude-sonnet-3-7", "tagline": "Extended-thinking update to Sonnet 3.5 with visible reasoning toggle.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "64K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2024-11", "license": "Proprietary", "pricing": { "input_per_mtok": "$3.00", "output_per_mtok": "$15.00" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "62.3%", "MMLU": "86.1%", "GPQA Diamond": "68.3%" }, "links": { "blog": "https://www.anthropic.com/news/claude-3-7-sonnet" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "DeepSeek-V3", "company": "DeepSeek", "release_date": "2024-12-26", "model_type": "LLM", "parameters": "671B", "key_features": [ "Mixture of Experts architecture", "Cost-effective training", "Open source release", "Strong reasoning capabilities" ], "performance_metrics": { "MMLU": "88.5%", "HumanEval": "90.2%", "MATH": "76.1%", "MMLU-Pro": "75.2%", "GPQA Diamond": "55.7%" }, "availability": "Open Source", "announcement_url": "https://github.com/deepseek-ai/DeepSeek-V3", "description": "DeepSeek's most advanced open-source model with MoE architecture", "category": "Major Release", "slug": "deepseek-deepseek-v3", "tagline": "DeepSeek's breakthrough open-weight MoE rivaling GPT-4-class quality.", "architecture": "Sparse MoE (37B active / 671B total)", "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "8K", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2024-07", "license": "DeepSeek License (open weights)", "pricing": { "input_per_mtok": "$0.27", "output_per_mtok": "$1.10" }, "providers": [ "DeepSeek API", "Hugging Face", "Together AI" ], "rate_limits": null, "benchmarks": { "MMLU": "88.5%", "HumanEval": "90.6%", "MATH": "61.6%", "MMLU-Pro": "75.2%", "GPQA Diamond": "55.7%", "LiveCodeBench": "35.9%", "MATH-500": "88.7%", "AIME 2025": "25.3%", "HLE": "3.6%", "SciCode": "35.4%", "TAU2-bench": "22.8%", "TerminalBench-Hard": "6.8%", "IF-Bench": "34.8%", "LiveCodeBench Reasoning": "29.0%" }, "links": { "repo": "https://github.com/deepseek-ai/DeepSeek-V3", "paper": "https://arxiv.org/abs/2412.19437" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemini 2.0 Flash", "company": "Google", "release_date": "2024-12-11", "model_type": "Multimodal", "parameters": "~175B", "key_features": [ "Native multimodal generation", "Real-time API", "Agentic capabilities", "Enhanced speed" ], "performance_metrics": { "MMLU": "85.8%", "HumanEval": "71.9%", "MATH": "58.8%", "MMLU-Pro": "77.9%", "GPQA Diamond": "62.3%" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/", "description": "Google's next-generation model with native multimodal capabilities", "category": "Major Release", "slug": "google-gemini-2-0-flash", "tagline": "First Gemini 2 model — fast, cheap, multimodal, with tool use native.", "architecture": "Dense multimodal transformer", "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "8K", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text", "image", "audio" ], "knowledge_cutoff": "2024-08", "license": "Proprietary", "pricing": { "input_per_mtok": "$0.10", "output_per_mtok": "$0.40", "cache_read_per_mtok": "$0.03" }, "providers": [ "Google AI Studio", "Vertex AI", "Gemini API" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "78.2%", "MMMU": "70.7%", "LiveCodeBench": "21.0%", "GPQA Diamond": "63.6%", "HumanEval": "90.7%", "MATH-500": "91.1%", "AIME 2025": "30.0%", "HLE": "4.7%", "SciCode": "34.0%", "TAU2-bench": "29.5%", "TerminalBench-Hard": "3.8%", "IF-Bench": "40.2%", "LiveCodeBench Reasoning": "28.3%" }, "links": { "blog": "https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Grok-2", "company": "xAI", "release_date": "2024-08-13", "model_type": "LLM", "parameters": "~314B", "key_features": [ "Real-time information access", "Multimodal understanding", "X platform integration", "Conversational AI" ], "performance_metrics": { "MMLU": "84.0%", "HumanEval": "74.1%", "MATH": "56.0%", "MMLU-Pro": "70.9%", "GPQA Diamond": "51.0%" }, "availability": "Platform Exclusive", "announcement_url": "https://x.ai/blog/grok-2", "description": "xAI's flagship model with real-time web access and multimodal capabilities", "category": "Major Release", "slug": "xai-grok-2", "tagline": "Elon's second-gen Grok — real-time X/Twitter data access", "architecture": "Transformer", "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": null, "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "Real-time (X feed)", "license": "Proprietary", "pricing": { "input": "$2/M", "output": "$10/M", "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [ "xAI API", "x.com (Grok)" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "70.9%", "GPQA Diamond": "51.0%", "HumanEval": "86.3%", "LiveCodeBench": "26.7%", "MATH-500": "77.8%", "AIME 2025": "13.3%", "HLE": "3.8%", "SciCode": "28.5%" }, "links": { "blog": "https://x.ai/blog/grok-2", "docs": "https://docs.x.ai" }, "predecessor_slug": "xai-grok-1-5", "changelog": [ "Vision input", "Real-time X data", "Improved reasoning" ], "launch_coverage": [] }, { "name": "Claude 3.5 Sonnet", "company": "Anthropic", "release_date": "2024-06-20", "model_type": "LLM", "parameters": "~175B", "key_features": [ "200K context window", "Improved coding capabilities", "Enhanced reasoning", "Vision capabilities" ], "performance_metrics": { "MMLU": "88.7%", "HumanEval": "92%", "MATH": "71.1%", "MMLU-Pro": "77.2%", "SWE-bench Verified": "49.0%", "GPQA Diamond": "59.9%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-3-5-sonnet", "description": "Anthropic's most intelligent model with significantly improved capabilities", "category": "Major Release", "slug": "anthropic-claude-3-5-sonnet", "tagline": "The mid-2024 Sonnet release that set the SOTA bar for coding and agents.", "architecture": "Dense Transformer (proprietary)", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "8K", "input_modalities": [ "text", "image", "PDF" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2024-04", "license": "Proprietary", "pricing": { "input_per_mtok": "$3.00", "output_per_mtok": "$15.00", "cache_read_per_mtok": "$0.30" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "SWE-bench Verified": "49.0%", "MMLU": "88.7%", "HumanEval": "89.9%", "GPQA Diamond": "56.0%", "MMLU-Pro": "75.1%", "LiveCodeBench": "38.1%", "MATH-500": "69.5%", "AIME 2025": "9.7%", "HLE": "3.7%", "SciCode": "31.6%" }, "links": { "blog": "https://www.anthropic.com/news/claude-3-5-sonnet" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude 3 Opus", "company": "Anthropic", "release_date": "2024-03-04", "model_type": "LLM", "parameters": "~175B", "key_features": [ "200K context window", "Advanced reasoning", "Multimodal capabilities", "Constitutional AI training" ], "performance_metrics": { "MMLU": "86.8%", "HumanEval": "84.9%", "MATH": "60.1%", "MMLU-Pro": "69.6%", "GPQA Diamond": "48.9%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-3-family", "description": "Most capable model in the Claude 3 family with near-human performance on complex tasks", "category": "Major Release", "slug": "anthropic-claude-3-opus", "tagline": "Anthropic's most powerful pre-Claude 4 model — tops GPT-4 on reasoning", "architecture": "Transformer", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "4096 tokens", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-08", "license": "Proprietary", "pricing": { "input": "$15/M", "output": "$75/M", "input_per_mtok": "$18.75", "output_per_mtok": "$75.00", "cache_read_per_mtok": "$1.50" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "86.8%", "HumanEval": "84.8%", "GSM8K": "95.0%", "MMLU-Pro": "69.6%", "GPQA Diamond": "48.9%", "LiveCodeBench": "27.9%", "MATH-500": "64.1%", "AIME 2025": "3.3%", "HLE": "3.1%", "SciCode": "23.3%" }, "links": { "blog": "https://www.anthropic.com/news/claude-3-family", "docs": "https://docs.anthropic.com" }, "predecessor_slug": "anthropic-claude-2-1", "changelog": [ "200K context", "Vision input", "+15% MMLU vs Claude 2", "Tool use" ], "launch_coverage": [] }, { "name": "Claude 3 Sonnet", "company": "Anthropic", "release_date": "2024-03-04", "model_type": "LLM", "parameters": "~175B", "key_features": [ "200K context window", "Balanced capability and speed", "Multimodal input", "Strong reasoning" ], "performance_metrics": { "MMLU": "79.0%", "HumanEval": "73.0%", "MATH": "40.5%", "MMLU-Pro": "57.9%", "GPQA Diamond": "40.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-3-family", "description": "Balanced model offering strong performance with faster response times", "category": "Major Release", "slug": "anthropic-claude-3-sonnet", "tagline": "Balanced Claude 3 variant — best price/performance in the family", "architecture": "Transformer", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "4096 tokens", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-08", "license": "Proprietary", "pricing": { "input": "$3/M", "output": "$15/M", "input_per_mtok": "$3.00", "output_per_mtok": "$15.00" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "79.0%", "HumanEval": "71.3%", "MMLU-Pro": "57.9%", "GPQA Diamond": "40.0%", "LiveCodeBench": "17.5%", "MATH-500": "41.4%", "AIME 2025": "4.7%", "HLE": "3.8%", "SciCode": "22.9%" }, "links": { "blog": "https://www.anthropic.com/news/claude-3-family", "docs": "https://docs.anthropic.com" }, "predecessor_slug": "anthropic-claude-2", "changelog": [], "launch_coverage": [] }, { "name": "Claude 3 Haiku", "company": "Anthropic", "release_date": "2024-03-04", "model_type": "LLM", "parameters": "~25B", "key_features": [ "200K context window", "Fastest response times", "Multimodal input", "Cost-effective" ], "performance_metrics": { "MMLU": "75.2%", "HumanEval": "75.9%", "MATH": "38.9%", "GPQA Diamond": "37.4%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-3-family", "description": "Fastest and most compact model in the Claude 3 family", "category": "Major Release", "slug": "anthropic-claude-3-haiku", "tagline": "Fastest and cheapest Claude 3 — sub-second latency at $0.25/M", "architecture": "Transformer", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": "4096 tokens", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-08", "license": "Proprietary", "pricing": { "input": "$0.25/M", "output": "$1.25/M", "input_per_mtok": "$0.25", "output_per_mtok": "$1.25" }, "providers": [ "Anthropic API", "AWS Bedrock", "Google Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "75.2%", "HumanEval": "75.7%", "GPQA Diamond": "37.4%", "LiveCodeBench": "15.4%", "MATH-500": "39.4%", "AIME 2025": "1.0%", "HLE": "3.9%", "SciCode": "18.6%", "TAU2-bench": "21.1%", "TerminalBench-Hard": "0.8%", "IF-Bench": "36.1%", "LiveCodeBench Reasoning": "21.0%" }, "links": { "blog": "https://www.anthropic.com/news/claude-3-family", "docs": "https://docs.anthropic.com" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Mistral Large", "company": "Mistral", "release_date": "2024-02-26", "model_type": "LLM", "parameters": "~70B", "key_features": [ "32K context window", "Multilingual capabilities", "Function calling", "JSON mode" ], "performance_metrics": { "MMLU": "81.2%", "HumanEval": "45%", "HellaSwag": "89.2%", "MMLU-Pro": "51.5%", "GPQA Diamond": "35.1%" }, "availability": "API Only", "announcement_url": "https://mistral.ai/news/mistral-large/", "description": "Top-tier reasoning model with strong multilingual capabilities", "category": "Major Release", "slug": "mistral-mistral-large", "tagline": null, "architecture": null, "training_tokens": null, "context_window": "32K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": null, "pricing": { "input_per_mtok": "$4.00", "output_per_mtok": "$12.00" }, "providers": [], "rate_limits": null, "benchmarks": { "MMLU-Pro": "51.5%", "GPQA Diamond": "35.1%", "HumanEval": "70.6%", "LiveCodeBench": "17.8%", "MATH-500": "52.7%", "AIME 2025": "0.0%", "HLE": "3.4%", "SciCode": "20.8%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemini 1.5 Pro", "company": "Google", "release_date": "2024-02-15", "model_type": "Multimodal", "parameters": "~175B", "key_features": [ "1M token context window", "Multimodal understanding", "Video analysis", "Audio processing" ], "performance_metrics": { "MMLU": "85.9%", "HumanEval": "71.9%", "MATH": "58.5%", "MMLU-Pro": "75.0%", "GPQA Diamond": "58.9%" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/", "description": "Google's next-generation model with breakthrough long context capabilities", "category": "Major Release", "slug": "google-gemini-1-5-pro", "tagline": "Google's first 1M-context model — multimodal needle-in-haystack champion", "architecture": "MoE Transformer", "training_tokens": null, "context_window": "1M tokens", "max_output_tokens": "8192 tokens", "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-11", "license": "Proprietary", "pricing": { "input (≤128K)": "$1.25/M", "input (>128K)": "$2.50/M", "output": "$5.00/M", "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [ "Google AI Studio", "Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "81.9%", "HumanEval": "83.4%", "MATH": "58.5%", "MMLU-Pro": "65.7%", "GPQA Diamond": "37.1%", "LiveCodeBench": "24.4%", "MATH-500": "67.3%", "AIME 2025": "8.0%", "HLE": "3.9%", "SciCode": "27.4%" }, "links": { "blog": "https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024", "docs": "https://ai.google.dev" }, "predecessor_slug": "google-gemini-pro", "changelog": [ "1M token context", "Multi-hour video understanding", "MoE architecture" ], "launch_coverage": [] }, { "name": "text-embedding-3-large", "company": "OpenAI", "release_date": "2024-01-25", "model_type": "Embedding", "parameters": "~7B", "key_features": [ "3072 embedding dimensions", "Improved retrieval performance", "Reduced hallucinations", "Multi-language support" ], "performance_metrics": { "MTEB Score": "64.6%", "Dimensions": "3072", "Languages": "100+" }, "availability": "API Only", "announcement_url": "https://openai.com/blog/new-embedding-models-and-api-updates", "description": "OpenAI's most powerful text embedding model", "category": "Major Release", "slug": "openai-text-embedding-3-large", "tagline": "OpenAI's best embedding model — 3× cheaper than ada-002 with better MTEB", "architecture": "Transformer encoder", "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "embeddings" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": { "input": "$0.13/M" }, "providers": [ "OpenAI API", "Azure OpenAI" ], "rate_limits": null, "benchmarks": { "MTEB avg": "64.6%" }, "links": { "blog": "https://openai.com/blog/new-embedding-models-and-api-updates", "docs": "https://platform.openai.com/docs/guides/embeddings" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GPT-4 Turbo", "company": "OpenAI", "release_date": "2024-01-25", "model_type": "LLM", "parameters": "~175B", "key_features": [ "128K context window", "Improved instruction following", "Enhanced reasoning capabilities", "Reduced hallucinations" ], "performance_metrics": { "MMLU": "86.4%", "HumanEval": "67%", "HellaSwag": "95.3%", "MMLU-Pro": "69.4%" }, "availability": "API Only", "announcement_url": "https://openai.com/blog/new-models-and-developer-products-announced-at-devday", "description": "Latest iteration of GPT-4 with improved performance and longer context window", "category": "Major Release", "slug": "openai-gpt-4-turbo", "tagline": "GPT-4 with 128K context and knowledge through April 2023 — 3× cheaper than GPT-4", "architecture": "Transformer", "training_tokens": null, "context_window": "128K tokens", "max_output_tokens": "4096 tokens", "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-04", "license": "Proprietary", "pricing": { "input": "$10/M", "output": "$30/M", "input_per_mtok": "$10.00", "output_per_mtok": "$30.00" }, "providers": [ "OpenAI API", "Azure OpenAI" ], "rate_limits": null, "benchmarks": { "MMLU-Pro": "69.4%", "HumanEval": "91.8%", "LiveCodeBench": "29.1%", "MATH-500": "73.7%", "AIME 2025": "15.0%", "HLE": "3.3%", "SciCode": "31.9%" }, "links": { "blog": "https://openai.com/blog/new-models-and-developer-products-announced-at-devday", "docs": "https://platform.openai.com/docs" }, "predecessor_slug": "openai-gpt-4", "changelog": [ "128K context (8× increase)", "Updated knowledge cutoff", "3× cheaper than GPT-4" ], "launch_coverage": [] }, { "name": "Grok-1", "company": "xAI", "release_date": "2023-12-07", "model_type": "LLM", "parameters": "~314B", "key_features": [ "Real-time information", "Conversational interface", "X platform integration", "Uncensored responses" ], "performance_metrics": { "MMLU": "73.0%", "HumanEval": "63.2%", "GSM8K": "62.9%" }, "availability": "Platform Exclusive", "announcement_url": "https://x.ai/blog/grok", "description": "xAI's first major language model with real-time internet access", "category": "Major Release", "slug": "xai-grok-1", "tagline": "xAI's open-source release — 314B MoE, first frontier model fully open-sourced", "architecture": "MoE Transformer", "training_tokens": null, "context_window": "8K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-10-01", "license": "Apache 2.0", "pricing": { "self-hosted": "Free", "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [ "Self-hosted (HuggingFace)" ], "rate_limits": null, "benchmarks": {}, "links": { "repo": "https://github.com/xai-org/grok-1", "blog": "https://x.ai/blog/grok-os" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "AlphaCode 2", "company": "Google DeepMind", "release_date": "2023-12-06", "model_type": "Code", "parameters": "~340B", "key_features": [ "Advanced code generation", "Competitive programming", "Multi-language support", "Problem decomposition" ], "performance_metrics": { "Codeforces Rating": "1747", "Problem Solving": "85th percentile", "Language Support": "10+ languages" }, "availability": "Research Only", "announcement_url": "https://deepmind.google/discover/blog/competitive-programming-with-alphacode/", "description": "DeepMind's advanced code generation system for competitive programming", "category": "Research", "slug": "google-deepmind-alphacode-2", "tagline": "DeepMind's coding specialist — top 15% of competitive programmers", "architecture": "Transformer (Gemini-based)", "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [ "text", "code" ], "output_modalities": [ "code" ], "knowledge_cutoff": null, "license": "Proprietary (research)", "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": { "Codeforces percentile": "Top 15%" }, "links": { "paper": "https://storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Gemini Ultra", "company": "Google", "release_date": "2023-12-06", "model_type": "Multimodal", "parameters": "~540B", "key_features": [ "Multimodal reasoning", "Text, image, audio, video understanding", "Advanced mathematical reasoning", "Code generation" ], "performance_metrics": { "MMLU": "90.0%", "HumanEval": "74.4%", "MATH": "53.2%" }, "availability": "Limited Access", "announcement_url": "https://deepmind.google/technologies/gemini/", "description": "Google's most capable multimodal AI model", "category": "Major Release", "slug": "google-gemini-ultra", "tagline": "Google's first model to beat GPT-4 on MMLU — 90%+ with CoT", "architecture": "MoE Transformer", "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [ "text", "image", "audio", "video" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-06", "license": "Proprietary", "pricing": { "Gemini Advanced": "$19.99/month" }, "providers": [ "Google One AI Premium", "Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "90.0%", "HumanEval": "74.4%", "MATH": "53.2%" }, "links": { "blog": "https://blog.google/technology/ai/google-gemini-ai" }, "predecessor_slug": "google-palm-2", "changelog": [ "First model to exceed human expert on MMLU", "Native multimodal", "32K context" ], "launch_coverage": [] }, { "name": "Gemini Pro", "company": "Google", "release_date": "2023-12-06", "model_type": "Multimodal", "parameters": "~175B", "key_features": [ "Multimodal capabilities", "32K context window", "Fast inference", "Scalable deployment" ], "performance_metrics": { "MMLU": "83.7%", "HumanEval": "67.7%", "MATH": "32.6%" }, "availability": "API Only", "announcement_url": "https://deepmind.google/technologies/gemini/", "description": "Google's balanced model for wide range of tasks", "category": "Major Release", "slug": "google-gemini-pro", "tagline": "Google's workhorse Gemini model — free tier in Google AI Studio", "architecture": "Transformer", "training_tokens": null, "context_window": "32K tokens", "max_output_tokens": null, "input_modalities": [ "text", "image" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-06", "license": "Proprietary", "pricing": { "input": "Free (AI Studio)", "Vertex AI": "$0.50/M" }, "providers": [ "Google AI Studio", "Vertex AI" ], "rate_limits": null, "benchmarks": { "MMLU": "79.1%", "HumanEval": "67.7%" }, "links": { "blog": "https://blog.google/technology/ai/google-gemini-ai", "docs": "https://ai.google.dev" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude 2.1", "company": "Anthropic", "release_date": "2023-11-21", "model_type": "LLM", "parameters": "~175B", "key_features": [ "200K context window", "Reduced hallucination rates", "Enhanced accuracy", "Tool use capabilities" ], "performance_metrics": { "MMLU": "88.0%", "HumanEval": "71.2%", "MATH": "71.1%", "MMLU-Pro": "49.5%", "GPQA Diamond": "31.9%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-2-1", "description": "Significant improvements in accuracy and honesty over Claude 2", "category": "Update", "slug": "anthropic-claude-2-1", "tagline": "Claude 2 update — 200K context and reduced hallucinations", "architecture": "Transformer", "training_tokens": null, "context_window": "200K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-01", "license": "Proprietary", "pricing": { "input": "$8/M", "output": "$24/M", "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [ "Anthropic API", "AWS Bedrock" ], "rate_limits": null, "benchmarks": { "MMLU": "73.1%", "HumanEval": "15.9%", "MMLU-Pro": "49.5%", "GPQA Diamond": "31.9%", "LiveCodeBench": "19.5%", "MATH-500": "37.4%", "AIME 2025": "3.3%", "HLE": "4.2%", "SciCode": "18.4%" }, "links": {}, "predecessor_slug": "anthropic-claude-2", "changelog": [ "200K context (2× Claude 2)", "50% fewer hallucinations", "Tool use beta" ], "launch_coverage": [] }, { "name": "Whisper v3", "company": "OpenAI", "release_date": "2023-11-06", "model_type": "Audio", "parameters": "~1.55B", "key_features": [ "Multilingual speech recognition", "99 language support", "Robust noise handling", "Real-time transcription" ], "performance_metrics": { "WER English": "5.1%", "Language Coverage": "99 languages", "Real-time Factor": "0.8x" }, "availability": "Open Source", "announcement_url": "https://github.com/openai/whisper", "description": "OpenAI's multilingual speech recognition system", "category": "Major Release", "slug": "openai-whisper-v3", "tagline": "State-of-the-art open speech recognition — 99 languages, open weights", "architecture": "Transformer encoder-decoder", "training_tokens": null, "context_window": null, "max_output_tokens": null, "input_modalities": [ "audio" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "MIT", "pricing": { "OpenAI API": "$0.006/minute", "self-hosted": "Free" }, "providers": [ "OpenAI API", "Self-hosted (HuggingFace)", "Groq" ], "rate_limits": null, "benchmarks": { "WER (English)": "~2.7%" }, "links": { "repo": "https://github.com/openai/whisper", "blog": "https://openai.com/research/whisper" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Code Llama 34B", "company": "Meta", "release_date": "2023-08-24", "model_type": "Code", "parameters": "34B", "key_features": [ "Code generation", "Code completion", "Multiple programming languages", "Large context window" ], "performance_metrics": { "HumanEval": "48.4%", "MBPP": "55.0%", "MultiPL-E": "45.9%" }, "availability": "Open Source", "announcement_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "description": "Specialized model for code generation built on Llama 2", "category": "Major Release", "slug": "meta-code-llama-34b", "tagline": "Meta's code-specialized open model — top open-source coding at launch", "architecture": "Transformer (Llama 2 fine-tune)", "training_tokens": null, "context_window": "100K tokens", "max_output_tokens": null, "input_modalities": [ "text", "code" ], "output_modalities": [ "code", "text" ], "knowledge_cutoff": "2023-01", "license": "Llama 2 Community License", "pricing": { "self-hosted": "Free" }, "providers": [ "Self-hosted", "Together AI", "Fireworks AI" ], "rate_limits": null, "benchmarks": { "HumanEval": "48.8%" }, "links": { "repo": "https://github.com/facebookresearch/codellama", "paper": "https://arxiv.org/abs/2308.12950" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Llama 2 70B", "company": "Meta", "release_date": "2023-07-18", "model_type": "LLM", "parameters": "70B", "key_features": [ "Open source", "Commercial license", "Improved safety", "Enhanced performance" ], "performance_metrics": { "MMLU": "68.9%", "HumanEval": "29.9%", "MATH": "13.5%" }, "availability": "Open Source", "announcement_url": "https://ai.meta.com/blog/llama-2/", "description": "Meta's open-source large language model with commercial license", "category": "Major Release", "slug": "meta-llama-2-70b", "tagline": "Meta's open-weight landmark — 70B that matched GPT-3.5 and ignited open AI", "architecture": "Transformer", "training_tokens": "2T tokens", "context_window": "4K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-01", "license": "Llama 2 Community License", "pricing": { "self-hosted": "Free", "Together AI": "$0.90/M" }, "providers": [ "Self-hosted", "Together AI", "AWS Bedrock", "Azure AI" ], "rate_limits": null, "benchmarks": { "MMLU": "68.9%", "HumanEval": "29.9%" }, "links": { "repo": "https://github.com/facebookresearch/llama", "paper": "https://arxiv.org/abs/2307.09288" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Claude 2", "company": "Anthropic", "release_date": "2023-07-11", "model_type": "LLM", "parameters": "~175B", "key_features": [ "100K context window", "Improved safety", "Enhanced reasoning", "Better code generation" ], "performance_metrics": { "MMLU": "78.5%", "HumanEval": "71.2%", "MATH": "88.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/claude-2", "description": "Significant improvement over Claude 1 with enhanced capabilities", "category": "Major Release", "slug": "anthropic-claude-2", "tagline": "Claude's first major leap — 100K context and better at instructions", "architecture": "Transformer", "training_tokens": null, "context_window": "100K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-01", "license": "Proprietary", "pricing": { "input": "$8/M", "output": "$24/M" }, "providers": [ "Anthropic API", "AWS Bedrock" ], "rate_limits": null, "benchmarks": { "MMLU": "78.5%", "HumanEval": "71.2%" }, "links": {}, "predecessor_slug": "anthropic-claude-1-3", "changelog": [ "100K context (10× Claude 1)", "Improved reasoning", "Reduced refusals" ], "launch_coverage": [] }, { "name": "PaLM 2", "company": "Google", "release_date": "2023-05-10", "model_type": "LLM", "parameters": "~340B", "key_features": [ "Multilingual capabilities", "Reasoning improvements", "Coding abilities", "Multiple model sizes" ], "performance_metrics": { "MMLU": "86.4%", "HumanEval": "77.2%", "MATH": "34.3%" }, "availability": "API Only", "announcement_url": "https://blog.google/technology/ai/google-palm-2-ai-large-language-model/", "description": "Google's improved large language model powering Bard and other services", "category": "Major Release", "slug": "google-palm-2", "tagline": "Google's multilingual flagship — powers Bard 2023, 100+ languages", "architecture": "Transformer", "training_tokens": null, "context_window": "8K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2023-02", "license": "Proprietary", "pricing": { "PaLM API (Bison)": "$0.50/M", "input_per_mtok": "$0.00", "output_per_mtok": "$0.00" }, "providers": [ "Google Cloud Vertex AI", "Google AI Studio" ], "rate_limits": null, "benchmarks": { "MMLU": "78.3%", "MATH": "34.3%" }, "links": { "paper": "https://arxiv.org/abs/2305.10403", "blog": "https://blog.google/technology/ai/google-palm-2-ai-large-language-model" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GPT-4", "company": "OpenAI", "release_date": "2023-03-14", "model_type": "LLM", "parameters": "~175B", "key_features": [ "8K context window", "Multimodal capabilities", "Enhanced reasoning", "Improved factual accuracy" ], "performance_metrics": { "MMLU": "86.4%", "HumanEval": "67.0%", "MATH": "42.5%" }, "availability": "API Only", "announcement_url": "https://openai.com/research/gpt-4", "description": "OpenAI's most advanced system producing safer and more useful responses", "category": "Major Release", "slug": "openai-gpt-4", "tagline": "The model that changed everything — GPT-4 set the standard for capable AI", "architecture": "Transformer (reported MoE)", "training_tokens": null, "context_window": "8K tokens (32K with gpt-4-32k)", "max_output_tokens": "4096 tokens", "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2021-09", "license": "Proprietary", "pricing": { "input": "$30/M", "output": "$60/M", "input_per_mtok": "$30.00", "output_per_mtok": "$60.00" }, "providers": [ "OpenAI API", "Azure OpenAI" ], "rate_limits": null, "benchmarks": { "MMLU": "86.4%", "HumanEval": "67.0%", "MATH": "52.9%", "Bar exam": "~90th percentile" }, "links": { "technical_report": "https://arxiv.org/abs/2303.08774", "blog": "https://openai.com/gpt-4" }, "predecessor_slug": "openai-chatgpt-gpt-3-5-turbo", "changelog": [ "Passed bar exam (top 10%)", "Vision input (GPT-4V)", "Multimodal" ], "launch_coverage": [] }, { "name": "Claude 1.3", "company": "Anthropic", "release_date": "2023-03-14", "model_type": "LLM", "parameters": "~52B", "key_features": [ "Constitutional AI", "Helpful and harmless", "Long conversations", "Improved reasoning" ], "performance_metrics": { "MMLU": "77.0%", "HumanEval": "56.0%", "MATH": "36.0%" }, "availability": "API Only", "announcement_url": "https://www.anthropic.com/news/introducing-claude", "description": "Anthropic's AI assistant built using Constitutional AI methods", "category": "Major Release", "slug": "anthropic-claude-1-3", "tagline": "Anthropic's first public model — 100K context ahead of its time", "architecture": "Transformer", "training_tokens": null, "context_window": "100K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2022-12", "license": "Proprietary", "pricing": { "input": "$8/M", "output": "$24/M" }, "providers": [ "Anthropic API" ], "rate_limits": null, "benchmarks": { "MMLU": "75.0%" }, "links": {}, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "ChatGPT (GPT-3.5 Turbo)", "company": "OpenAI", "release_date": "2022-11-30", "model_type": "LLM", "parameters": "~175B", "key_features": [ "Conversational interface", "Fine-tuned for chat", "RLHF training", "Fast response times" ], "performance_metrics": { "MMLU": "70.0%", "HumanEval": "48.1%", "MATH": "34.1%" }, "availability": "Free + Paid Tiers", "announcement_url": "https://openai.com/blog/chatgpt", "description": "Conversational AI that sparked mainstream adoption of large language models", "category": "Major Release", "slug": "openai-chatgpt-gpt-3-5-turbo", "tagline": "The product that launched the AI era — 100M users in 2 months", "architecture": "Transformer (RLHF fine-tune of GPT-3.5)", "training_tokens": null, "context_window": "16K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2021-09", "license": "Proprietary", "pricing": { "input": "$0.50/M", "output": "$1.50/M" }, "providers": [ "OpenAI API", "Azure OpenAI" ], "rate_limits": null, "benchmarks": { "MMLU": "70.0%", "HumanEval": "48.1%" }, "links": { "blog": "https://openai.com/blog/chatgpt", "docs": "https://platform.openai.com/docs" }, "predecessor_slug": "openai-instructgpt", "changelog": [ "Conversational interface", "RLHF alignment", "Faster and cheaper than GPT-4" ], "launch_coverage": [] }, { "name": "PaLM", "company": "Google", "release_date": "2022-04-04", "model_type": "LLM", "parameters": "540B", "key_features": [ "Large parameter count", "Few-shot learning", "Reasoning capabilities", "Code generation" ], "performance_metrics": { "MMLU": "69.3%", "HumanEval": "26.2%", "MATH": "8.8%" }, "availability": "Research Only", "announcement_url": "https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html", "description": "Google's 540-billion parameter language model demonstrating breakthrough capabilities", "category": "Research", "slug": "google-palm", "tagline": "Google's 540B pathways model — first to demonstrate chain-of-thought at scale", "architecture": "Pathways Transformer", "training_tokens": "780B tokens", "context_window": null, "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": null, "license": "Proprietary (research)", "pricing": {}, "providers": [], "rate_limits": null, "benchmarks": { "MMLU": "69.3%", "BIG-bench": "58.1%" }, "links": { "paper": "https://arxiv.org/abs/2204.02311", "blog": "https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "Codex", "company": "OpenAI", "release_date": "2021-08-10", "model_type": "Code", "parameters": "~12B", "key_features": [ "Code generation", "Natural language to code", "Multiple programming languages", "GitHub Copilot integration" ], "performance_metrics": { "HumanEval": "28.8%", "MBPP": "59.6%", "APPS": "25.0%" }, "availability": "API Only", "announcement_url": "https://openai.com/blog/openai-codex", "description": "AI system that translates natural language to code", "category": "Major Release", "slug": "openai-codex", "tagline": "GPT-3 trained on code — the engine behind GitHub Copilot v1", "architecture": "Transformer (GPT-3 fine-tune)", "training_tokens": null, "context_window": "4K tokens", "max_output_tokens": null, "input_modalities": [ "text", "code" ], "output_modalities": [ "code" ], "knowledge_cutoff": null, "license": "Proprietary", "pricing": {}, "providers": [ "OpenAI API (deprecated)", "GitHub Copilot" ], "rate_limits": null, "benchmarks": { "HumanEval": "28.8%" }, "links": { "paper": "https://arxiv.org/abs/2107.03374", "blog": "https://openai.com/blog/openai-codex" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] }, { "name": "GPT-3", "company": "OpenAI", "release_date": "2020-06-11", "model_type": "LLM", "parameters": "175B", "key_features": [ "175 billion parameters", "Few-shot learning", "Text generation", "Multiple capabilities" ], "performance_metrics": { "MMLU": "43.9%", "HumanEval": "0.0%", "MATH": "5.2%" }, "availability": "API Only", "announcement_url": "https://openai.com/blog/gpt-3-apps", "description": "Breakthrough large language model that demonstrated emergent capabilities", "category": "Major Release", "slug": "openai-gpt-3", "tagline": "The model that showed scaling works — 175B parameters, few-shot learning pioneer", "architecture": "Transformer", "training_tokens": "300B tokens", "context_window": "2K tokens", "max_output_tokens": null, "input_modalities": [ "text" ], "output_modalities": [ "text" ], "knowledge_cutoff": "2019-10", "license": "Proprietary (API)", "pricing": { "Davinci": "$0.02/1K tokens" }, "providers": [ "OpenAI API" ], "rate_limits": null, "benchmarks": { "MMLU": "43.9%" }, "links": { "paper": "https://arxiv.org/abs/2005.14165", "blog": "https://openai.com/blog/openai-api" }, "predecessor_slug": null, "changelog": [], "launch_coverage": [] } ]