New OpenAI benchmark tests how well models reason complex drug discovery tasks
MedChemBench tests AI reasoning across 99 chemistry problems, exposing gaps in drug property prediction. Key Takeaways MedChemBench evaluates model performance across 99 medicinal chemistry problems drawn from 117 papers and patents. GPT-6 Astra earned 95.4% of available credit for molecular structure reconstruction at the xhigh reasoning setting. The benchmark evaluates reasoning from historical research [...]









