SciLaws-Bench - Real Symbolic Validity: leaderboard
SciLaws-Bench symbolic validity: whether the discovered law is correct on the real-world task set, independent of how novel its form is. Higher is better. Source: arxiv.org/abs/2609.01552.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 81.84 | #33 |
| 2 | DeepSeek V4 Pro | 81.3 | #112 |
| 3 | Claude Opus 4.8 | 80.24 | #39 |
| 4 | GPT-5 Mini | 77.03 | #193 |
| 5 | Gemini 3.5 Flash | 74.63 | #66 |
| 6 | GLM-5.2 | 74.19 | #80 |
| 7 | Qwen 3.7 Max | 72.66 | #85 |
| 8 | GPT-5.4 Mini | 72.45 | #228 |
| 9 | GPT-4o Mini | 56.02 | #598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scilaws-bench-real-symbolic-validity · How It Works · Data refreshed daily, snapshot 2026-11-05.