ChemRxn-Bench Judge - Direct Products: leaderboard
AppendixTable7 secondary judge: atom-level validity/semantic conditions separate from main rule-based exact-molecule scoring. RP judge precision,RR rule recall,RF1 per-figure judge macroF1; parentheses stereochemistry-blind. Weighted Overall judge composite. Same603figure manifest and zero-credit failure policy; source method-specific results.
Metric: Judge score (%). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 83.79 | #16 |
| 2 | Kimi K3 | 67.54 | #35 |
| 3 | GPT-5.6 Sol | 66.94 | #22 |
| 4 | Seed 2.1 Pro | 66.92 | #73 |
| 5 | Gemini 3.1 Pro (Preview) | 59.97 | #64 |
| 6 | GLM-5.3 Flash | 56.57 | #77 |
| 7 | Qwen3.8-Flash-Next | 49.89 | #76 |
| 8 | DeepSeek V4.1 Flash | 27.34 | #37 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=chemrxn-bench-judge-direct-products · How It Works · Data refreshed daily, snapshot 2026-11-05.