MolViBench (Agent Collaboration): leaderboard

MolViBench (2026), a benchmark of 358 RDKit molecular coding tasks, run under Agent Collaboration: a Coder and a Tester instance of the same model exchange test plans and pass or fail feedback for up to three rounds. The score is Pass@1 against the reference outputs, in percent; higher is better.

Metric: Pass@1 (%) over all 358 tasks under Agent Collaboration: a Coder and a Tester instance of the same model exchange test plans and PASS or FAIL feedback for up to 3 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 July 2028. 3 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro36#145
2Claude Opus 4.634.1#113
3DeepSeek V3.2Thinking34.1#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-agent-collaboration · How It Works · Data refreshed daily, snapshot 2026-11-21.