Vellum - GPQA: leaderboard
The GPQA Diamond column of Vellum's public LLM leaderboard, covering graduate-level multiple choice questions in biology, physics and chemistry that were written and validated by subject experts. Vellum states its figures come from model providers together with independently run evaluations, not from one harness it controls. Higher accuracy is better. Source: arxiv.org/abs/2311.12022.
Metric: Accuracy (%). Source: www.vellum.ai. Saturation forecast: Estimated already saturated. 72 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5 | 96.2 | #57 |
| 2 | GPT-6 Astra | 96 | #6 |
| 3 | GPT-5.6 Sol | 94.6 | #19 |
| 4 | Gemini 3.1 Pro (Preview) | 94.3 | #58 |
| 5 | Claude Opus 4.7 | 94.2 | #49 |
| 6 | Claude Fable 5 | 94.1 | #16 |
| 7 | Claude Mythos 5 | 94.1 | #2 |
| 8 | GPT-5.5 | 93.6 | #31 |
| 9 | Claude Opus 4.8 | 93.6 | #37 |
| 10 | Kimi K3 | 93.5 | #30 |
| 11 | MiniMax-M3 | 93 | #136 |
| 12 | GPT-5.6 Terra | 92.9 | #41 |
| 13 | GPT-5.4 | 92.8 | #79 |
| 14 | GPT-5.2 | 92.4 | #110 |
| 15 | GPT-5.6 Luna | 92.3 | #68 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #8 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=vellum-gpqa · How It Works · Data refreshed daily, snapshot 2026-10-20.