SciDesignBench - Optimization: leaderboard
Metric: One-turn design optimization: the model receives a starting design with its measured properties and must modify it to meet new targets (success also requires changing the design); mean success rate (%) over the 10 optimization shared-core domains, 20 tasks each, averaged over K = 3 attempts; higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 34.5 | #60 |
| 2 | Gemini 3.1 Pro (Preview) | 31.3 | #54 |
| 3 | Claude Sonnet 4.6 | 31.2 | #85 |
| 4 | Claude Sonnet 4.5 | 30.5 | #138 |
| 5 | GPT-5.2 | 29 | #105 |
| 6 | Gemini 2.0 Flash | 22.2 | #331 |
| 7 | GPT-4o | 17.8 | #333 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=scidesignbench-optimization · How It Works · Data refreshed daily, snapshot 2026-10-11.