OpenAI launches BioBasicsBench to test AI agents on everyday wet-lab data

OpenAI on 6 November 2026 introduced BioBasicsBench, a benchmark developed with Ligo, a biotech company focused on enzyme design, to measure how well AI agents interpret raw experimental data from everyday wet-lab work — including failed experiments, ambiguous images and conflicting observations.

What the benchmark tests

The set has 100 tasks across five areas: protein methods, imaging and cytometry, quantification and assays, nucleic acid methods, and molecular cloning. Inputs include gels and blots (31 tasks), protocols and notes (27), tables and instrument exports (20), bench photographs (18), microscopy (14), instrument screens and plots (13), sequencing files (12), protocol videos (7) and agar plates (7); categories overlap. Scientists wrote tasks from their own experiments with expert rubrics; each task had 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.

OpenAI evaluated seven of its models across 37 model-and-effort settings, with five attempts per task, for 18,500 results. At the xhigh reasoning setting, GPT-6 Astra earned a mean 66.7% of available credit (95% CI 60.3–73.1%, about 4,704 output tokens per task), versus 43.8% for GPT-5.6 Sol (95% CI 37.6–50.0%, about 9,099 tokens). Other xhigh scores included GPT-6.1 Sol 62.0%, GPT-6 Luna 38.3%, GPT-5.5 35.4%, GPT-5.4 33.2% and GPT-5.2 31.2%.

Limits

These are OpenAI’s own evaluations of its own models; the company compared only its lineup, and independent reproduction is not reported here. OpenAI says higher scores “do not yet mean models can make reliable judgments in the lab,” citing remaining errors on controls, colony counting (about a 12% overcount on one plate) and experimental context. The benchmark measures routine analysis, not autonomous lab work or clinical decision-making.