100-task OpenAI benchmark assesses AI agents' ability to handle routine wet-lab science workflows

OpenAI has introduced BioBasicsBench, a new evaluation framework comprising 100 tasks that measure how effectively AI agents interpret raw experimental data and manage everyday wet-lab science procedures, according to an official advisory from the company.reported

The figures

The benchmark contains exactly 100 tasks, as stated in the OpenAI advisory.reported No other numerical metrics, such as pass rates or latency figures, are provided in the source material.

Behind the figures

The assessment measures an agent’s ability to interpret raw experimental data and handle routine laboratory workflows. The source material does not specify the underlying model architectures, the specific datasets used for these tasks, or the technical methodology for scoring the 100 items.

What changes, what does not

This release adds a specific metric for evaluating AI agents in wet-lab contexts. The advisory does not indicate any changes to existing OpenAI models or APIs, nor does it state that previous benchmarks are obsolete. The focus remains on measuring current capabilities rather than altering them.

What builders can do with it

Developers can use BioBasicsBench to evaluate how well their AI agents interpret raw experimental data and handle everyday wet-lab science tasks. The 100-task structure allows for granular testing of data interpretation skills.reported Builders can identify gaps in an agent’s ability to process laboratory workflows. The benchmark provides a standardised way to compare agent performance in scientific data handling. No integration code or API endpoints are documented in the advisory.