MolViBench (Incremental Repair): leaderboard
MolViBench (2026), a benchmark of 358 RDKit molecular coding tasks, run under Incremental Repair: the generated code is executed on test molecules and runtime errors are returned to the model for up to three repair rounds. The score is Pass@1 against the reference outputs, in percent; higher is better.
Metric: Pass@1 (%) over all 358 tasks under Incremental Repair: the code is run on ZINC250K test molecules and runtime errors are fed back for up to 3 repair rounds, RDKit molecular coding tasks answered by generated Python code, official APIs, temperature 0, at most 4,096 output tokens; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 3 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Opus 4.6 | Thinking | 39.7 | #113 |
| 2 | Gemini 3 Pro | 36 | #145 | |
| 3 | DeepSeek V3.2 | Thinking | 34.6 | #419 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=molvibench-incremental-repair · How It Works · Data refreshed daily, snapshot 2026-11-21.