SciUniverse - Optimize a single reaction: leaderboard
Task family in C5R's SciUniverse Level 1 (2026-09-24): maximize the yield of one amide coupling with 24 simulated experiments in three plates of eight (recorded AIChemEco outcomes), then nominate a condition. Passes when its confirmation yield closes at least 90% of the gap from a random-search baseline to the task-constrained oracle. 16 tasks, 4 attempts each. Pass@1 (%), higher is better.
Metric: Pass@1 (%). Source: c5r.net. Saturation forecast: Not forecast. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.8 Flash (High) | 26.6 | #36 (Gemini 3.8 Flash) |
| 2 | Claude Opus 5 (xHigh) | 20.3 | #20 (Claude Opus 5) |
| 3 | Grok 4.6 (xHigh) | 14.1 | #44 (Grok 4.6) |
| 4 | Claude Fable 5.1 (xHigh) | 7.8 | #14 (Claude Fable 5.1) |
| 5 | GPT-5.6 Sol (xHigh) | 4.7 | #29 (GPT-5.6 Sol) |
| 6 | GPT-6 Astra (xHigh) | 4.7 | #6 (GPT-6 Astra) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sciuniverse-optimize-a-single-reaction · How It Works · Data refreshed daily, snapshot 2026-11-12.