SciLaws-Bench - Parallel Symbolic Score: leaderboard
SciLaws-Bench parallel worlds: the same discovery tasks with the constants altered, so a recalled law does not score. Higher is better. Source: arxiv.org/abs/2609.01552.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around May 2028. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 58.26 | #33 |
| 2 | GLM-5.2 | 52.12 | #79 |
| 3 | Gemini 3.5 Flash | 50.42 | #67 |
| 4 | Qwen 3.7 Max | 50.42 | #86 |
| 5 | Claude Opus 4.8 | 50 | #40 |
| 6 | DeepSeek V4 Pro | 49.79 | #112 |
| 7 | GPT-5 Mini | 44.7 | #192 |
| 8 | GPT-5.4 Mini | 42.58 | #229 |
| 9 | GPT-4o Mini | 33.47 | #599 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scilaws-bench-parallel-symbolic-score · How It Works · Data refreshed daily, snapshot 2026-11-06.