Key Info

OpenAI and Ligo, a biotech company focused on enzyme design, released BioBasicsBench, a benchmark built from 100 tasks drawn from real experiments to measure how AI agents handle routine analyses in wet-lab science.

Highlights

  • The benchmark tests whether agents can interpret raw lab data, identify errors, and recognize when evidence does not support a conclusion.
  • Tasks are sourced from real experiments rather than synthetic scenarios.
  • OpenAI frames the benchmark as a way to pinpoint where models need improvement and to track gains for scientific progress, biosafety, and biosecurity.