URSA - Expert-2026 ChemCensor: leaderboard
Metric: Average ChemCensor plausibility score (out of 5) per reaction step along the best route per target, averaged over the 100 URSA-expert-2026 novel targets; URSA-minor-1.1.0-U2 configuration with ChemCensor v1.2.0 and the U2 USPTO precedent database; default reasoning effort; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 1.62 |
| 2 | Gemini 3.1 Pro (Preview) | 1.46 |
| 3 | Kimi K2.5 | 1.44 |
| 4 | Claude Opus 4.8 | 1.3 |
| 5 | Claude Opus 4.7 | 1.24 |
| 6 | Grok 4.1 | 1.22 |
| 7 | Claude Opus 4.5 | 1.03 |
| 8 | Claude Opus 4.6 | 1.02 |
| 9 | Qwen 3.5 397B A17B | 1 |
| 10 | GLM-5 | 1 |
| 11 | Claude Sonnet 4.6 | 0.99 |
| 12 | Claude Sonnet 4.5 | 0.98 |
| 13 | Grok 4.3 | 0.94 |
| 14 | DeepSeek V3.2 | 0.85 |
| 15 | GPT-5.4 | 0.71 |
Interactive version: theaggregate.ai/benchmark?slug=ursa-expert-2026-chemcensor · How It Works · Data refreshed daily, snapshot 2026-09-29.