Graduate-level science and reasoning benchmark score.
GPQA, often reported as GPQA Diamond in model leaderboards, is a graduate-level Google-proof Q&A benchmark. It focuses on expert science reasoning where strong retrieval alone is not enough.
Test type: Expert multiple-choice science Q&A, usually evaluated with exact option extraction.
Top models ranked by GPQA.
| Rank | Model | Creator | Value | Speed | Blended Price |
|---|---|---|---|---|---|
| #1 | Gemini 3.1 Pro Preview | 94.1% | 124.7 tok/s | $4.50/M | |
| #2 |
| OpenAI |
| 93.5% |
| 69 tok/s |
| $11.25/M |
| #3 | GPT-5.5 (high) | OpenAI | 93.2% | 61.6 tok/s | $11.25/M |
| #4 | MiniMax-M3 | MiniMax | 92.9% | 45.6 tok/s | $0.525/M |
| #5 | Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) | Anthropic | 92.6% | n/a | $20.00/M |
| #6 | GPT-5.5 (medium) | OpenAI | 92.6% | 58.7 tok/s | $11.25/M |
| #7 | Qwen3.7 Max | Alibaba | 92.3% | 186.5 tok/s | $3.75/M |
| #8 | Gemini 3.5 Flash (high) | 92.2% | 203.3 tok/s | $3.38/M |
| #9 | Gemini 3.5 Flash (medium) | 92.1% | 210.1 tok/s | $3.38/M |
| #10 | Claude Opus 4.8 (Adaptive Reasoning, Max Effort) | Anthropic | 92.0% | 67.8 tok/s | $10.00/M |
| #11 | GPT-5.4 (xhigh) | OpenAI | 92.0% | 75.5 tok/s | $5.63/M |
| #12 | GPT-5.3 Codex (xhigh) | OpenAI | 91.5% | 84.5 tok/s | $4.81/M |
| #13 | Claude Opus 4.7 (Adaptive Reasoning, Max Effort) | Anthropic | 91.4% | 53.8 tok/s | $10.00/M |
| #14 | Grok 4.20 0309 v2 (Reasoning) | xAI | 91.1% | 168.7 tok/s | $3.00/M |
| #15 | Kimi K2.6 | Kimi | 91.1% | 41.6 tok/s | $1.71/M |
| #16 | GPT-5.5 (low) | OpenAI | 91.0% | 66.4 tok/s | $11.25/M |
| #17 | Gemini 3 Pro Preview (high) | 90.8% | n/a | $4.50/M |
| #18 | DeepSeek V4 Pro (Reasoning, High Effort) | DeepSeek | 90.5% | 65.7 tok/s | $0.544/M |
| #19 | GPT-5.2 (xhigh) | OpenAI | 90.3% | 71 tok/s | $4.81/M |
| #20 | Grok 4.3 (high) | xAI | 90.1% | 159.7 tok/s | $1.56/M |
| #21 | Qwen3.7 Plus | Alibaba | 90.0% | 53.6 tok/s | $0.590/M |
| #22 | GPT-5.2 Codex (xhigh) | OpenAI | 89.9% | 105.3 tok/s | $4.81/M |
| #23 | Gemini 3 Flash Preview (Reasoning) | 89.8% | 172.8 tok/s | $1.13/M |
| #24 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | Anthropic | 89.6% | 47.3 tok/s | $10.94/M |
| #25 | DeepSeek V4 Flash (Reasoning, Max Effort) | DeepSeek | 89.4% | 98.3 tok/s | $0.175/M |
| #26 | Qwen3.5 397B A17B (Reasoning) | Alibaba | 89.3% | 51.8 tok/s | $1.35/M |
| #27 | Grok 4.3 (medium) | xAI | 89.0% | 136.9 tok/s | $1.56/M |
| #28 | DeepSeek V4 Pro (Reasoning, Max Effort) | DeepSeek | 88.8% | 61.6 tok/s | $0.544/M |
| #29 | Qwen3.6 Max Preview | Alibaba | 88.8% | 40.9 tok/s | $2.93/M |
| #30 | Gemini 3 Pro Preview (low) | 88.7% | n/a | $4.50/M |
| #31 | Claude Opus 4.7 (Non-reasoning, High Effort) | Anthropic | 88.5% | 46 tok/s | $10.00/M |
| #32 | Grok 4.20 0309 (Reasoning) | xAI | 88.5% | 166.5 tok/s | $3.00/M |
| #33 | Muse Spark | Meta | 88.4% | n/a | - |
| #34 | Qwen3.6 Plus | Alibaba | 88.2% | 52.8 tok/s | $1.13/M |
| #35 | Kimi K2.5 (Reasoning) | Kimi | 87.9% | 31.7 tok/s | $1.19/M |
| #36 | Grok 4 | xAI | 87.7% | n/a | $11.00/M |
| #37 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) | Anthropic | 87.5% | 63.2 tok/s | $6.00/M |
| #38 | GPT-5.4 mini (xhigh) | OpenAI | 87.5% | 178.8 tok/s | $1.69/M |
| #39 | MiniMax-M2.7 | MiniMax | 87.4% | 75 tok/s | $0.525/M |
| #40 | GPT-5.1 (high) | OpenAI | 87.3% | 121.2 tok/s | $3.44/M |
| #41 | DeepSeek V3.2 Speciale | DeepSeek | 87.1% | n/a | - |
| #42 | GPT-5.4 (low) | OpenAI | 87.1% | 63.6 tok/s | $5.63/M |
| #43 | MiMo-V2-Pro | Xiaomi | 87.0% | 42.5 tok/s | $1.50/M |
| #44 | GLM-5.1 (Reasoning) | Z AI | 86.8% | 46.8 tok/s | $2.15/M |
| #45 | DeepSeek V4 Flash (Reasoning, High Effort) | DeepSeek | 86.7% | n/a | $0.175/M |
| #46 | Hy3-preview (Reasoning) | Tencent | 86.7% | 96 tok/s | $0.200/M |
| #47 | Claude Opus 4.5 (Reasoning) | Anthropic | 86.6% | 53.5 tok/s | $10.94/M |
| #48 | MiMo-V2.5-Pro | Xiaomi | 86.6% | 43.3 tok/s | $0.544/M |
| #49 | GPT-5.2 (medium) | OpenAI | 86.4% | n/a | $4.81/M |
| #50 | Qwen3 Max Thinking | Alibaba | 86.1% | n/a | $2.40/M |
| #51 | Qwen3.5 397B A17B (Non-reasoning) | Alibaba | 86.1% | 53.1 tok/s | $1.35/M |
| #52 | GPT-5.1 Codex (high) | OpenAI | 86.0% | 182.1 tok/s | $3.44/M |
| #53 | GLM-4.7 (Reasoning) | Z AI | 85.9% | 79.2 tok/s | $1.00/M |
| #54 | Qwen3.5 27B (Reasoning) | Alibaba | 85.8% | 82.8 tok/s | $0.825/M |
| #55 | Gemma 4 31B (Reasoning) | 85.7% | 34.8 tok/s | - |
| #56 | Qwen3.5 122B A10B (Reasoning) | Alibaba | 85.7% | 143.6 tok/s | $1.10/M |
| #57 | Ring-2.6-1T | InclusionAI | 85.7% | 122.1 tok/s | $0.850/M |
| #58 | KAT Coder Pro V2 | KwaiKAT | 85.5% | 118.1 tok/s | $0.525/M |
| #59 | MiMo-V2-Omni-0327 | Xiaomi | 85.5% | 85.6 tok/s | $0.800/M |
| #60 | GPT-5 (high) | OpenAI | 85.4% | 111.1 tok/s | $3.44/M |
| #61 | Grok 4.1 Fast (Reasoning) | xAI | 85.3% | n/a | - |
| #62 | MiMo-V2.5 | Xiaomi | 84.9% | 77.4 tok/s | $0.175/M |
| #63 | Nanbeige4.1-3B | Nanbeige | 84.9% | n/a | - |
| #64 | MiniMax-M2.5 | MiniMax | 84.8% | 202.9 tok/s | $0.525/M |
| #65 | GLM-5-Turbo | Z AI | 84.7% | n/a | - |
| #66 | Grok 4 Fast (Reasoning) | xAI | 84.7% | n/a | $0.275/M |
| #67 | GPT-5.5 Instant (May 2026) | OpenAI | 84.6% | n/a | $11.25/M |
| #68 | MiMo-V2-Flash (Reasoning) | Xiaomi | 84.6% | 129.5 tok/s | $0.150/M |
| #69 | o3-pro | OpenAI | 84.5% | 22.7 tok/s | $35.00/M |
| #70 | Qwen3.5 35B A3B (Reasoning) | Alibaba | 84.5% | 130.4 tok/s | $0.688/M |
| #71 | Gemini 2.5 Pro | 84.4% | 132 tok/s | $3.44/M |
| #72 | Grok 4.3 (low) | xAI | 84.3% | 148.4 tok/s | $1.56/M |
| #73 | GPT-5 (medium) | OpenAI | 84.2% | 85.6 tok/s | $3.44/M |
| #74 | Qwen3.5 27B (Non-reasoning) | Alibaba | 84.2% | 90.6 tok/s | $0.875/M |
| #75 | Qwen3.6 27B (Reasoning) | Alibaba | 84.2% | 54.7 tok/s | $1.35/M |
| #76 | Qwen3.6 35B A3B (Reasoning) | Alibaba | 84.1% | 159.9 tok/s | $0.557/M |
| #77 | Claude Opus 4.6 (Non-reasoning, High Effort) | Anthropic | 84.0% | 40.9 tok/s | $10.94/M |
| #78 | DeepSeek V3.2 (Reasoning) | DeepSeek | 84.0% | n/a | $0.337/M |
| #79 | GLM-5.1 (Non-reasoning) | Z AI | 83.9% | 45.6 tok/s | $2.15/M |
| #80 | Kimi K2 Thinking | Kimi | 83.8% | 131.1 tok/s | $1.08/M |
