Vector Eval - GPQA-D: leaderboard
Vector Institute Inspect Evals run of GPQA Diamond, testing graduate-level science reasoning on questions designed to challenge domain experts.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Coder Next | 75.63 | #326 |
| 2 | O1 | 75.51 | #183 |
| 3 | O3 Mini | 73.65 | #272 |
| 4 | DeepSeek R1 | 70.45 | #251 |
| 5 | Claude 3.5 Sonnet | 60.98 | #350 |
| 6 | Gemini 1.5 Pro | 57.83 | #445 |
| 7 | GPT-4o | 51.01 | #343 |
| 8 | Mistral Large 2 (Jul) | 47.35 | #484 |
| 9 | Llama 3.1 70B Instruct | 43.18 | #561 |
| 10 | Gemini 1.5 Flash | 40.4 | #618 |
| 11 | GPT-4o Mini | 38.38 | #599 |
| 12 | Command-R+ | 31.94 | #891 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vector-eval-gpqa-d · How It Works · Data refreshed daily, snapshot 2026-10-20.