The Best LLMs on GPQA Diamond
Which model answers the most graduate-level science questions?
84 of the 128 buyable models publish a GPQA Diamond score. It measures graduate-level reasoning, not coding, so it is reported here on its own and nowhere folded into a composite.
Method: Published GPQA Diamond score, descending, exactly as reported. GPQA Diamond is a graduate-level science question set, so it measures reasoning rather than software work, and it is never mixed into the coding quality score. Pricing as of July 2026. 23 rows below carry a score measured at a non-default reasoning tier and are marked as such; every affected model is named on the hub. Read the full method.
| Rank | Model | GPQA Diamond | AA index | Blended $/1M | Context |
|---|---|---|---|---|---|
| #1 | Kimi K3Moonshot | 93.5 | 57.1 | $6.00 | 1.05M |
| #2 | Grok 4.5xAI | 93.1 | 53.8 | $3.00 | 500K |
| #3 | MiniMax M3MiniMax | 92.9 | 44.4 | $1.05 | 1M |
| #4 | Gemini 3.6 FlashGoogle | 92.8 | 50.1 | $1.50 | 1.05M |
| #5 | GPT-5.6 SolOpenAI | 92.6 | 53.6 | $8.00 | 1.05M |
| #6 | GPT-5.5OpenAI | 92.6 | 50.4 | $11.25 | 1.05M |
| #7 | Claude Fable 5Anthropic | 92.6 | 59.9 | $20.00 | 1M |
| #8 | Qwen3.7 MaxAlibaba | 92.3 | 46.0 | $3.75 | 1M |
| #9 | Gemini 3.5 FlashGoogle | 92.1 | 45.4 | $3.38 | 1.05M |
| #10 | Claude Opus 4.8Anthropic | 92.0 | 55.7 | $10.00 | 1M |
| #11 | Claude Opus 5Anthropic | 91.9 | 56.3 | $10.00 | 1M |
| #12 | GPT-5.3 CodexOpenAI | 91.5 | 44.3 | $4.81 | 400K |
| #13 | Claude Opus 4.7Anthropic· non-default tier | 91.4 | 53.5 | $10.00 | 1M |
| #14 | Grok 4.20xAI· non-default tier | 91.1 | 37.0 | $1.56 | 1M |
| #15 | Kimi K2.6Moonshot | 91.1 | 44.2 | $1.71 | 262K |
| #16 | Claude Sonnet 5Anthropic· non-default tier | 91.1 | 53.4 | $4.00 | 1M |
| #17 | DeepSeek V4-ProDeepSeek· non-default tier | 90.5 | 43.1 | $0.990 | 1M |
| #18 | Qwen3.7 PlusAlibaba | 90.0 | 39.0 | $0.700 | 1M |
| #19 | Muse Spark 1.1Meta | 89.8 | 50.6 | $2.00 | 1M |
| #20 | Kimi K2.7 CodeMoonshot | 89.6 | 41.9 | $1.71 | 262K |
| #21 | Claude Opus 4.6Anthropic· non-default tier | 89.6 | 43.7 | $10.00 | 1M |
| #22 | GLM-5.2Zhipu· non-default tier | 89.5 | 51.1 | $2.15 | 1M |
| #23 | Nex-N2-ProNex AGI | 89.2 | 41.0 | $1.00 | 262K |
| #24 | Grok 4.3xAI | 89.0 | 36.0 | $1.56 | 1M |
| #25 | Qwen3.6-Max-PreviewAlibaba | 88.8 | 40.0 | $2.92 | 262K |
| #26 | Qwen3.6-PlusAlibaba | 88.2 | 39.6 | $1.13 | 1M |
| #27 | Kimi K2.5Moonshot· non-default tier | 87.9 | 35.4 | $1.20 | 262K |
| #28 | Claude Sonnet 4.6Anthropic· non-default tier | 87.5 | 47.2 | $6.00 | 1M |
| #29 | MiniMax M2.7MiniMax | 87.4 | 38.1 | $0.525 | 205K |
| #30 | GPT-5.1OpenAI· non-default tier | 87.3 | 36.9 | $3.44 | 400K |
| #31 | GPT-5.6 TerraOpenAI | 87.2 | 45.6 | $4.50 | 1.05M |
| #32 | GPT-5.4OpenAI· non-default tier | 87.1 | 39.1 | $5.63 | 1.05M |
| #33 | MiMo-V2-ProXiaomi | 87.0 | 40.3 | $1.50 | 1.05M |
| #34 | GLM-5.1Zhipu· non-default tier | 86.8 | 40.2 | $2.15 | 200K |
| #35 | DeepSeek V4-FlashDeepSeek· non-default tier | 86.7 | 37.5 | $0.330 | 1M |
| #36 | MiMo-V2.5-ProXiaomi | 86.6 | 42.2 | $0.544 | 1.05M |
| #37 | Claude Opus 4.5Anthropic· non-default tier | 86.6 | 40.8 | $10.00 | 200K |
| #38 | GPT-5.2OpenAI | 86.4 | 38.0 | $4.81 | 400K |
| #39 | GPT-5.6 LunaOpenAI | 85.9 | 38.1 | $0.450 | 1.05M |
| #40 | GLM-4.7Zhipu· non-default tier | 85.9 | 33.7 | $1.00 | 200K |
| #41 | Qwen3.5-27BAlibaba· non-default tier | 85.8 | 33.8 | $0.825 | 262K |
| #42 | MiniMax M2.5MiniMax | 84.8 | 33.7 | $0.525 | 128K |
| #43 | GLM-5 TurboZhipu | 84.7 | 38.1 | $1.90 | 200K |
| #44 | o3-proOpenAI | 84.5 | 32.5 | $35.00 | 200K |
| #45 | Gemini 2.5 ProGoogle | 84.4 | 25.8 | $3.44 | 1.05M |
| #46 | GPT-5OpenAI | 84.2 | 33.7 | $3.44 | 400K |
| #47 | Gemini 3.5 Flash-LiteGoogle | 83.8 | 36.5 | $0.850 | 1.05M |
| #48 | o3OpenAI | 82.7 | 30.4 | $3.50 | 200K |
| #49 | Qwen3.5-Omni PlusAlibaba | 82.6 | 30.6 | $1.50 | 262K |
| #50 | GPT-5.4 MiniOpenAI | 82.3 | 29.8 | $1.69 | 400K |
| #51 | Gemini 3.1 Flash-LiteGoogle | 82.2 | 25.0 | $0.563 | 1.05M |
| #52 | GLM-5Zhipu· non-default tier | 82.0 | 39.5 | $1.55 | 200K |
| #53 | Step 3.7 FlashStepFun | 80.9 | 30.3 | $0.438 | 262K |
| #54 | Qwen3.5-9BAlibaba· non-default tier | 80.6 | 21.4 | $0.113 | 262K |
| #55 | GPT-5 MiniOpenAI | 80.3 | 30.9 | $0.688 | 400K |
| #56 | Gemini 2.5 FlashGoogle· non-default tier | 79.0 | 20.1 | $0.850 | 1.05M |
| #57 | o4 MiniOpenAI | 78.4 | 25.6 | $1.93 | 200K |
| #58 | LongCat-2.0Meituan | 78.0 | 33.5 | $1.30 | 1M |
| #59 | Mercury 2Inception Labs | 77.0 | 21.4 | $0.375 | 128K |
| #60 | Mistral Small 4Mistral· non-default tier | 76.9 | 19.6 | $0.263 | 256K |
| #61 | Nova 2.0 LiteAmazon | 76.8 | 19.0 | $0.850 | 1M |
| #62 | GPT-5.4 NanoOpenAI | 76.1 | 30.2 | $0.463 | 400K |
| #63 | Command ACohere | 76.1 | 22.5 | $4.38 | 128K |
| #64 | Command A+Cohere | 76.1 | 22.5 | $4.38 | 128K |
| #65 | Gemma 4 12BGoogle· non-default tier | 75.3 | 21.8 | $0.00 | 262K |
| #66 | o3 MiniOpenAI | 74.8 | 19.0 | $1.93 | 200K |
| #67 | Mistral Medium 3.5Mistral | 74.8 | 29.9 | $3.00 | 256K |
| #68 | o1OpenAI | 74.7 | 23.4 | $26.25 | 200K |
| #69 | Qwen3.5-Omni FlashAlibaba | 74.2 | 19.0 | $0.850 | 262K |
| #70 | Qwen3 Coder NextAlibaba | 73.7 | 21.1 | $0.290 | 262K |
| #71 | Mistral Large 3Mistral | 68.0 | 15.9 | $0.750 | 262K |
| #72 | Llama 4 MaverickMeta | 67.1 | 14.3 | $0.415 | 1.05M |
| #73 | GPT-5 NanoOpenAI | 67.0 | 19.0 | $0.138 | 400K |
| #74 | GPT-4.1OpenAI | 66.6 | 19.4 | $3.50 | 1.05M |
| #75 | GPT-4.1 MiniOpenAI | 66.4 | 14.8 | $0.700 | 1.05M |
| #76 | Gemini 2.5 Flash-LiteGoogle· non-default tier | 62.5 | 11.4 | $0.175 | 1.05M |
| #77 | Llama 4 ScoutMeta | 58.7 | 10.0 | $0.135 | 1.05M |
| #78 | GLM-4.7-FlashXZhipu· non-default tier | 58.1 | 22.9 | $0.152 | 200K |
| #79 | Sonar ProPerplexity | 57.8 | 9.3 | $6.00 | 128K |
| #80 | Ministral 3 14BMistral | 57.2 | 11.1 | $0.200 | 262K |
| #81 | GPT-4oOpenAI· non-default tier | 54.3 | 11.2 | $4.38 | 128K |
| #82 | GPT-4.1 NanoOpenAI | 51.2 | 9.6 | $0.175 | 1.05M |
| #83 | Ministral 3 8BMistral | 47.1 | 9.0 | $0.150 | 262K |
| #84 | GPT-4o MiniOpenAI | 42.6 | 6.9 | $0.263 | 128K |
Frequently asked
Nearby questions, answered elsewhere
All rankings
The hub and the methodThe Best LLMs by Benchmark ScoreThe Best Value LLMs per DollarThe Best LLMs by Context WindowThe Cheapest LLMs With a 1M-Token Context WindowThe Best LLMs on Terminal-Bench 2.1The Best LLMs on Terminal-Bench HardThe Best LLMs on SWE-bench VerifiedThe Best LLMs for Tool Use on tau2-bench