BioBasicsBench measures AI agents on routine wet-lab data
OpenAI and Ligo introduced BioBasicsBench, a 100-task benchmark for how well AI agents read raw wet-lab data; GPT-6 Astra posted the highest mean reward at 66.7%.
OpenAI and Ligo introduced BioBasicsBench, a benchmark that measures how well AI agents interpret raw experimental data from routine wet-lab work, meaning bench experiments run on physical samples rather than in software. The benchmark contains 100 tasks across five areas of routine wet-lab work.
Seven models were evaluated: GPT-5.2, GPT-5.4, GPT-5.5, GPT-5.6 Sol, GPT-6.1 Sol, GPT-6 Luna and GPT-6 Astra. The evaluation covers 37 model and reasoning-effort settings, with five attempts on each of the 100 tasks per setting, for a total of 18,500 results.
Why it matters
At the xhigh setting, which allows more reasoning per task, GPT-6 Astra earned an average of 66.7% of the available credit, compared with 43.8% for GPT-5.6 Sol. On one DNA-cloning control plate, GPT-6 Astra caught a wrong antibiotic label that GPT-5.6 Sol missed. The label “LB Kan50” identifies kanamycin, a different antibiotic from the ampicillin used in the experiment. GPT-6 Astra also correctly recognized a z-stack, a series of images taken at different depths.
What to do
Each task underwent an internal review by a Ligo scientist who had not written it, an external review by a PhD-level scientist working in a wet lab, and a final internal review. Before quoting a score, check both the review record behind the task set and the model and reasoning-effort setting the number came from.
Sources
Related reading
3 postsDiscussion
Spotted an error? Send corrections to [email protected] .