SEAL - SciPredict: leaderboard
405 questions from post-2025 empirical studies across physics, chemistry, and biology, testing models' ability to predict real scientific experimental outcomes.
Metric: Score. Source: scale.com. Source: benchmarklist.com. Saturation forecast: Around 2033. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 25.27 | #66 |
| 2 | Claude Opus 4.5 (20251101) | 23.05 | #74 |
| 3 | Claude Sonnet 4.5 | 22.55 | #142 |
| 4 | Gemini 3 Flash | 22.22 | #99 |
| 5 | Claude Opus 4.1 (20250805) | 22.22 | #132 |
| 6 | GPT-5.2 | 20.58 | #110 |
| 7 | O3 Mini (2025-01-31) | 19.84 | #240 |
| 8 | DeepSeek V3 | 19.18 | #321 |
| 9 | Llama 3.3 70B | 18.19 | #579 |
| 10 | Gemini 2.5 Pro | 17.04 | #152 |
| 11 | Qwen 3 32B | 17.04 | #443 |
| 12 | Qwen 3 235B A22B | 16.63 | #313 |
| 13 | Llama 3.1 8B | 14.65 | #1151 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-scipredict · How It Works · Data refreshed daily, snapshot 2026-10-20.