Benchmark RadarAI BENCHMARK PROFILE
ChemCoTBench-V2
ChemCoTBench-V2 evaluates chemical reasoning at the process level, with 5,620 samples across molecular understanding, editing, optimization, and reaction prediction. It checks intermediate steps using deterministic chemistry rules and reference traces.
- Released
- 2026-06-02
- Readiness
- Paper only
- Primary field
- Science & Research
Why it matters
Existing chemistry benchmarks only score final answers, missing violations in reasoning. ChemCoTBench-V2 provides auditable, verifiable process-level evaluation with three separate signals, enabling fine-grained model comparison.
Motivation
Large language models are increasingly used as chemistry assistants, yet most chemistry benchmarks still score only final answers.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.