AI Models, Software & Research

OpenAI launched the 99-question MedChemBench medicinal-chemistry benchmark; GPT-6.2 Astra scored 84.5% in ultra-high reasoning mode.

· ComputeLabs Research · from the November 20, 2026 edition

OpenAI said it introduced MedChemBench to assess reasoning relevant to medicinal-chemistry research. Its 99 questions, drawn from published research, test whether models can connect molecular structures with experimental evidence, explain results, compare compounds, and predict properties.

The reported scores in ultra-high reasoning mode were 84.5% for GPT-6.2 Astra, 83.3% for GPT-6 Astra, 82.0% for GPT-6.1 Sol, and 58.2% for GPT-5.6 Sol. These are benchmark results attributed to OpenAI through Financial_Express, rather than independently reproduced measurements in the supplied sources.

The report says GPT-6 Astra performed best on reconstructing molecular structures and interpreting experimental evidence, while compound screening and numerical-property prediction were more difficult for that model. The supplied account does not disclose scoring methodology, repeated-run variability, or validation in laboratory drug-development workflows.

Additional reporting

  • OpenAI
  • MedChemBench
  • GPT-6.2 Astra

All 19 stories from November 20, 2026