URSA Single-Step Retrosynthesis - Max ChemCensor: leaderboard
Metric: Average over the 100 URSA-expert-2026 novel targets of the highest ChemCensor score (0 to 5) among the model's predicted reactant sets for the target (15 independent samples per target, five-shot prompts drawn from USPTO-full training reactions, no tools, reasoning enabled at the default budget for models that support it; ChemCensor 0.5.2 scores each reaction from 0 (no precedent) to 5 by the substructural context of its reaction centre and functional groups found in USPTO-full precedents); higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 1.8 | #78 |
| 2 | Grok 4.1 (Thinking) | 1.75 | #218 (Grok 4.1) |
| 3 | Claude Sonnet 4.5 (Thinking) | 1.45 | #138 (Claude Sonnet 4.5) |
| 4 | Claude Opus 4.5 (Thinking) | 1.34 | #79 (Claude Opus 4.5) |
| 5 | Kimi K2 | 1.12 | #236 |
| 6 | GPT-5.2 (Thinking) | 0.85 | #105 (GPT-5.2) |
| 7 | GPT-5.1 (Thinking) | 0.73 | #131 (GPT-5.1) |
| 8 | Gemini 2.5 Flash | 0.66 | #237 |
| 9 | DeepSeek V3.2 (Thinking) | 0.39 | #198 (DeepSeek V3.2) |
| 10 | Qwen 3 14B | 0.01 | #524 |
| 11 | Qwen 3 8B | 0 | #667 |
| 12 | LFM2 2.6B | 0 | #1286 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=ursa-single-step-retrosynthesis-max-chemcensor · How It Works · Data refreshed daily, snapshot 2026-10-11.