OpenRouter GPQA Diamond: leaderboard
One item is a four-option graduate-level science question from the fixed GPQA Diamond set, sent to live provider endpoints through OpenRouter rather than run by the model's vendor. Scores aggregate repeated runs weighted by question count, with a minimum sample count per model and provider pair, and the headline row uses default routing where available. Higher is better.
Metric: Accuracy (%). Source: openrouter.ai. Saturation forecast: Estimated already saturated. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Argon | 96.5 | #9 |
| 2 | Gemini 3.8 Flash | 95.6 | #30 |
| 3 | Gemini 3.1 Pro (Preview) | 94.5 | #64 |
| 4 | GPT-6 Astra | 94.4 | #6 |
| 5 | GPT-6.1 Sol | 94.4 | #7 |
| 6 | GPT-5.6 Pro Sol | 94.1 | #21 |
| 7 | GPT-5.5 | 93.9 | #33 |
| 8 | Gemini 3.7 Flash | 93.9 | #24 |
| 9 | Kimi K3.1 Pro | 93.1 | #17 |
| 10 | Gemini 3.6 Flash | 92.8 | #57 |
| 11 | Claude Sonnet 5.5 | 92.8 | #14 |
| 12 | Grok 4.6 | 92.8 | #29 |
| 13 | GPT-5.6 Sol | 92.4 | #22 |
| 14 | Gemini 3.5 Flash | 92.3 | #66 |
| 15 | Kimi K3 | 91.9 | #35 |
Interactive version: theaggregate.ai/benchmark?slug=openrouter-gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-11-05.