AI chemist solves 20-year-old bottleneck using GPT‑5.4
openai and molecule.one used gpt‑5.4 to boost chan‑lam coupling yields by 52% in autonomous lab experiments, potentially unlocking faster drug discovery.
openai, in collaboration with molecule.one, connected gpt‑5.4 to the high-throughput marai lab to autonomously optimize synthetic chemistry reactions. focusing on chan‑lam coupling, the model identified primary sulfonamides as a high-value challenge and proposed using temo as an additive, which increased mean yields from 16.6% to 25.2% and moved 37.5% of reactions above the 30% yield threshold. human chemists validated these results at bench scale, confirming a twofold yield increase in most cases. this success demonstrates that ai agents can operate effectively in physical labs to overcome historical bottlenecks in drug synthesis, expanding the library of molecules available for pharmaceutical testing.