OpenAI and Ligo release BioBasicsBench — a test for AI agents in...

OpenAI and Ligo release BioBasicsBench — a test for AI agents in the 'wet' lab

OpenAI, in collaboration with biotech company Ligo, has introduced BioBasicsBench — a benchmark of 100 tasks involving routine 'wet' lab work where an AI agent interprets raw experimental data. GPT-6 Astra achieved the highest average score with partial credit at the maximum reasoning level — 66.7% (95% CI 60.3–73.1%).

Compiled by Sergey KostenchukPublished 2026-11-10Updated 2026-11-10

2026-11-10 Research OpenAI

🧪 BioBasicsBench: AI agents tested in the 'wet' lab

OpenAI and biotech company Ligo have released 100 tasks where an agent interprets raw data: gels and blots (31), protocols and records (27), tables and instrument exports (20), photos, microscopy, and video. The highest score at maximum reasoning was 66.7% for GPT-6 Astra; 62.0% for GPT-6.1 Sol and 43.8% for GPT-5.6 Sol.

🌍 The first public benchmark measures not biological knowledge, but the handling of raw data and experimental failures — a metric for 'lab' AI agents.

👤 The BioBasicsBench leaderboard is an indicator of the maturity of scientific AI agents, but 66.7% represents partial scores, there is no independent replication, and OpenAI has not disclosed the full rubrics, raw data, or code. For example, Astra rejected a cloning control with an incorrect label 'LB Kan50'.

Source 1: https://openai.com/index/measuring-how-ai-agents-handle-wet-lab-science/ Source 2: https://kingy.ai/blog/biobasicsbench-openai-wet-lab-ai-guide/

Sources