OpenAI launches wet-lab benchmark for AI agents
OpenAI and Ligo released BioBasicsBench, 100 routine wet-lab tasks built on raw experimental data, on which GPT-6 Astra earned 66.7% of available credit versus 43.8% for GPT-5.6 Sol.
OpenAI introduced BioBasicsBench, a benchmark that tests whether AI agents can interpret the raw, often messy evidence produced in everyday wet-lab work. It was developed with Ligo, a biotech company focused on enzyme design.
OpenAI argues that existing benchmarks built on figures selected for papers miss the failed experiments, ambiguous images and conflicting observations that working scientists troubleshoot. The 100 tasks cover five areas: protein methods, imaging and cytometry, quantification and assays, nucleic acid methods, and molecular cloning. Inputs include gel and blot images, plate photographs, microscopy stacks, sequencing reads, plate-reader exports and videos of bench work.
Scientists wrote the tasks from their own experimental data, each paired with an expert-written rubric. Every task passed an internal review by a Ligo scientist who had not written it, an external review by a PhD-level wet-lab scientist, and a final internal review. Rubric criteria award partial credit for conclusions, numerical results, reasoning, or recognizing that the evidence is insufficient.
OpenAI evaluated seven of its own models across 37 model and reasoning-effort settings, with five attempts per task, for 18,500 results. At the "xhigh" setting, GPT-6 Astra scored 66.7% against 43.8% for GPT-5.6 Sol. OpenAI says Astra's gains held even at low reasoning effort and suggests stronger native multimodal capabilities may be key.
In one example, GPT-5.6 Sol classified cells as dead based on a dye signal the detector could confuse with a second dye; Astra flagged the ambiguity and asked for better controls. OpenAI says image interpretation, control checking and use of experimental context still need work. Only OpenAI models were reported.