Intel / article

LifeSciBench

Source: openai.com
Date: Published Jun 17, 2026

Context

OpenAI's LifeSciBench evaluates responses to 750 research tasks using detailed scientific rubrics. In its reported results, GPT-Rosalind passed 36.1% overall, compared with GPT-5.5's 25.7%. A pass means reaching a 70% rubric threshold, so it does not mean every requirement was satisfied.

The results also show weaker performance on tasks with supporting files. GPT-Rosalind passed 45.1% of text-only tasks but 28.1% of tasks with artifacts or URLs. These are different task groups, not a controlled experiment removing attachments. OpenAI says models struggle to extract information from complex figures or large sequence files and integrate it into an answer. The is OpenAI-built and covers self-contained tasks, not live research.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Read the source openai.com