GPQA Diamond: leaderboard
Graduate-level Google-Proof Q&A benchmark evaluating models on questions that require deep reasoning.
Metric: Accuracy (%). Source: www.vals.ai. Status: saturated. 138 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 95.45 |
| 2 | GPT-5.6 Sol (Max) | 95.2 |
| 3 | Grok 4.6 (High) | 94.7 |
| 4 | Gemini 3.8 Flash (High) | 94.44 |
| 5 | Gemini 3.7 Flash (High) | 93.94 |
| 6 | Qwen 3.8 Max | 93.69 |
| 7 | Claude Opus 5 | 93.43 |
| 8 | Gemini 3.6 Flash (High) | 93.43 |
| 9 | Claude Fable 5.1 (Max) | 93.43 |
| 10 | GPT-5.5 (xHigh) | 93.18 |
| 11 | Claude Fable 5 (Max) | 93.18 |
| 12 | Kimi K3 | 92.93 |
| 13 | Grok 4.5 (High) | 92.93 |
| 14 | MiniMax-M3 | 92.68 |
| 15 | Gemini 3.5 Flash (High) | 92.68 |
Interactive version: theaggregate.ai/benchmark?slug=gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-09-05.