ZeroEval GPQA Diamond: leaderboard
This board mirrors the GPQA leaderboard that the ZeroEval API serves to LLM Stats, tagged there as the diamond subset. The items are expert-written four-option science questions in biology, physics and chemistry. Nearly all rows are accuracies self-reported by the model providers rather than independent runs, none is marked verified, and higher is better. Source: llm-stats.com/benchmarks/gpqa.
Metric: GPQA Diamond Score. Source: huggingface.co. Source: www.vals.ai. Saturation forecast: Estimated already saturated. 250 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 96 | #6 |
| 2 | GPT-5.6 Sol | 94.6 | #19 |
| 3 | Claude Mythos Preview | 94.6 | #4 |
| 4 | Gemini 3.1 Pro (Preview) | 94.3 | #58 |
| 5 | Claude Opus 4.7 | 94.2 | #49 |
| 6 | GPT-5.5 | 93.6 | #31 |
| 7 | Claude Opus 4.8 | 93.6 | #37 |
| 8 | Kimi K3 | 93.5 | #30 |
| 9 | GPT-5.2 Pro | 93.2 | #61 |
| 10 | Grok 4.5 | 93 | #48 |
| 11 | GPT-5.6 Terra | 92.9 | #41 |
| 12 | GPT-5.4 | 92.8 | #79 |
| 13 | Qwen 3.8 Max | 92.6 | #35 |
| 14 | GPT-5.2 | 92.4 | #110 |
| 15 | Qwen 3.7 Max | 92.4 | #75 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=zeroeval-gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-10-20.