OpenCompass Research - GPQA Diamond: leaderboard
OpenCompass research-table score on GPQA Diamond, the hardest subset of graduate-level science questions.
Metric: Score (%). Source: rank.opencompass.org.cn. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 91.5 | #66 |
| 2 | Qwen 3.5 397B A17B | 88.4 | #146 |
| 3 | Kimi K2.5 | 88.1 | #144 |
| 4 | GLM-4.7 | 86.9 | #192 |
| 5 | DeepSeek V3.2 Speciale | 86.7 | #128 |
| 6 | GPT-5 (2025-08-07) (High) | 86.4 | |
| 7 | GLM-5 | 85.3 | #141 |
| 8 | Gemini 2.5 Pro | 84.7 | #152 |
| 9 | DeepSeek V3.2 | 84.6 | #205 |
| 10 | MiniMax-M2.5 | 84.6 | #303 |
| 11 | Step 3.5 Flash | 83.7 | #270 |
| 12 | Kimi K2 (Thinking) | 82.7 | #243 (Kimi K2) |
| 13 | GPT-5 Mini (2025-08-07) (High) | 82.7 | #172 (GPT-5 Mini (2025-08-07)) |
| 14 | O3 (2025-04-16) | 82.3 | #121 |
| 15 | MiMo-V2-Flash | 82.1 | #372 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=opencompass-research-gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-10-20.