LiveCodeBench leaderboard
LiveCodeBench is a contamination-free coding benchmark composed of programming problems collected after model training cut-offs.
38 models ranked, highest score first.
| # | Model | Company | Score |
|---|---|---|---|
| 1 | GPT-5.2 | OpenAI | 88.9% |
| 2 | Claude Opus 4.5 | Anthropic | 87.1% |
| 3 | GPT-5.1 | OpenAI | 86.8% |
| 4 | DeepSeek V3.2 | DeepSeek | 86.2% |
| 5 | GPT-5.3 Codex | OpenAI | 84.2% |
| 6 | GPT-5.2 Codex | OpenAI | 80.4% |
| 7 | Gemini 3.1 Pro | 78.9% | |
| 8 | MiniMax-M3 | MiniMax | 74.0% |
| 9 | Claude Fable 5 | Anthropic | 70.0% |
| 10 | Gemini 2.5 Flash | 69.5% | |
| 11 | Qwen3.7 Max | Alibaba | 69.0% |
| 12 | Kimi K2 | Moonshot AI | 68.9% |
| 13 | Claude Opus 4.8 | Anthropic | 67.7% |
| 14 | Qwen3.7 Plus | Alibaba | 65.0% |
| 15 | Ring-2.6-1T | InclusionAI | 64.3% |
| 16 | Step 3.7 Flash | StepFun | 63.7% |
| 17 | Claude Sonnet 4.6 | Anthropic | 58.7% |
| 18 | GPT-5 | OpenAI | 55.8% |
| 19 | Gemma 4 12B | 55.3% | |
| 20 | JT-35B-Flash | China Mobile | 55.3% |
| 21 | Gemini 3.5 Flash | 53.3% | |
| 22 | Mistral Large 3 | Mistral | 46.5% |
| 23 | Claude 3.5 Sonnet | Anthropic | 38.1% |
| 24 | DeepSeek-V3 | DeepSeek | 35.9% |
| 25 | North Mini Code | Cohere | 32.3% |
| 26 | HyperNova 60B 2605 | Multiverse Computing | 31.7% |
| 27 | GPT-4 Turbo | OpenAI | 29.1% |
| 28 | Claude 3 Opus | Anthropic | 27.9% |
| 29 | Grok-2 | xAI | 26.7% |
| 30 | Gemini 1.5 Pro | 24.4% | |
| 31 | Gemini 2.0 Flash | 21.0% | |
| 32 | Claude 2.1 | Anthropic | 19.5% |
| 33 | Mistral Large | Mistral | 17.8% |
| 34 | Claude 3 Sonnet | Anthropic | 17.5% |
| 35 | Claude 3 Haiku | Anthropic | 15.4% |
| 36 | MiniCPM-V 4.6 1.3B | OpenBMB | 6.3% |
| 37 | MiniCPM5-1B | OpenBMB | 3.7% |
| 38 | LFM2.5-8B-A1B | Liquid AI | 0.0% |