AI Flash Report

GPQA Diamond leaderboard

GPQA Diamond is the hardest tier of the GPQA benchmark — graduate-level science questions designed to be Google-proof.

56 models ranked, highest score first.

GPQA Diamond leaderboard — 56 models ranked by score
# Model Company Score
1 GPT-5.5 OpenAI 93.5%
2 MiniMax-M3 MiniMax 92.9%
3 Claude Fable 5 Anthropic 92.6%
4 Qwen3.7 Max Alibaba 92.3%
5 Claude Opus 4.8 Anthropic 92.0%
6 GPT-5.3 Codex OpenAI 91.5%
7 GPT-5.2 OpenAI 90.3%
8 Grok 4.3 xAI 90.1%
9 Qwen3.7 Plus Alibaba 90.0%
10 DeepSeek V4 Flash DeepSeek 89.4%
11 DeepSeek V4 Pro DeepSeek 88.8%
12 Qwen3.6 Max Preview Alibaba 88.8%
13 GPT-5.1 OpenAI 87.3%
14 Hy3-preview Tencent 86.7%
15 MiMo-V2.5-Pro Xiaomi 86.6%
16 Claude Opus 4.5 Anthropic 86.6%
17 Ring-2.6-1T InclusionAI 85.7%
18 MiMo-V2.5 Xiaomi 84.9%
19 GPT-5.5 Instant OpenAI 84.6%
20 Qwen3.6 27B Alibaba 84.2%
21 Gemini 3.1 Pro Google 84.2%
22 DeepSeek V3.2 DeepSeek 84.0%
23 JT-35B-Flash China Mobile 82.9%
24 Gemini 3.5 Flash Google 82.8%
25 Step 3.7 Flash StepFun 80.9%
26 Claude Sonnet 4.6 Anthropic 79.7%
27 Claude Opus 4.1 Anthropic 79.1%
28 Gemini 2.5 Flash Google 79.0%
29 Gemini 3 Pro Google 78.5%
30 North Mini Code Cohere 75.7%
31 Gemma 4 12B Google 75.3%
32 Ling-2.6-1T InclusionAI 75.2%
33 Mistral Medium 3.5 Mistral 74.8%
34 Kimi K2 Moonshot AI 74.1%
35 Claude Sonnet 4 Anthropic 74.0%
36 HyperNova 60B 2605 Multiverse Computing 73.3%
37 Claude Sonnet 3.7 Anthropic 68.3%
38 Mistral Large 3 Mistral 68.0%
39 GPT-5 OpenAI 67.3%
40 Gemini 2.0 Flash Google 63.6%
41 Ling 2.6 Flash InclusionAI 59.3%
42 Claude 3.5 Sonnet Anthropic 56.0%
43 DeepSeek-V3 DeepSeek 55.7%
44 LFM2.5-8B-A1B Liquid AI 51.3%
45 Grok-2 xAI 51.0%
46 Claude 3 Opus Anthropic 48.9%
47 Granite 4.1 30B IBM 48.1%
48 Granite 4.1 8B IBM 43.3%
49 Claude 3 Sonnet Anthropic 40.0%
50 Claude 3 Haiku Anthropic 37.4%
51 Gemini 1.5 Pro Google 37.1%
52 Mistral Large Mistral 35.1%
53 Claude 2.1 Anthropic 31.9%
54 Granite 4.1 3B IBM 31.4%
55 MiniCPM-V 4.6 1.3B OpenBMB 30.5%
56 MiniCPM5-1B OpenBMB 27.8%