OpenAI Boosts AI for Drug Discovery

OpenAI is rolling out significant upgrades to its GPT-Rosalind AI model, specifically targeting the complex demands of life sciences research and drug discovery. The updated model, built on the foundation of OpenAI GPT-5.5, now boasts enhanced agentic coding, tool-use capabilities, and deeper intelligence in critical areas like medicinal chemistry and genomics.

This advancement aims to bridge the gap between AI-driven reasoning and the practical execution of scientific workflows. GPT-Rosalind's improvements are designed to synthesize data across various scales, from molecules to living systems, a crucial aspect of progress in the field.

Tackling Scientifically Valuable Tasks

To gauge its real-world impact, OpenAI developed LifeSciBench, a comprehensive benchmark judged by external experts. This evaluation covers six core life science workflow areas: evidence handling, analysis, design and optimization, scientific reasoning, validation, and communication.

In evaluations using LifeSciBench, the updated GPT-Rosalind demonstrated broad performance gains across tasks identified by both academic and industry professionals. This includes intricate medicinal chemistry queries and complex biological analyses.

Stronger Scientific Reasoning

In medicinal chemistry, GPT-Rosalind achieved industry-leading performance, as measured by MedChemBench. It reportedly outperformed GPT-5.5 by 27.5% to 25.1% while utilizing fewer tokens, showcasing improved multimodal synthesis and mechanistic reasoning.

For genomics and quantitative biology, evaluated via GeneBench, GPT-Rosalind showed a higher accuracy of 21.6% compared to GPT-5.5's 20.4%, using 31% fewer tokens. This long-horizon, end-to-end analysis capability extends to functional genomics, spatial transcriptomics, and proteomics.

Assisting Real-World Lab Work

A new evaluation, LabWorkBench, specifically tests GPT-Rosalind's ability to assist scientists with real-world lab protocols, from troubleshooting to optimization. The model scored 63.2% compared to GPT-5.5's 55.8% on this proprietary dataset, indicating significant gains in practical lab assistance and token efficiency.

From Reasoning to Executed Workflows

OpenAI has also introduced two plugins, Life Sciences Research and Life Sciences NGS Analysis, to provide a practical execution layer for repeatable scientific workflows. These plugins integrate evidence retrieval, biological interpretation, and bioinformatics execution, allowing researchers to connect external data with internal analyses.

Qualified enterprise users can now leverage GPT-Rosalind to power these plugins, enhancing the utility of Codex as a dynamic workbench. New interactive viewers for native biological file types, such as sequences and alignments, are also being integrated to keep scientists closely connected to the data during AI-driven reasoning.

The company is offering expanded access for trusted organizations globally through a research preview deployment structure, signaling a push towards deeper AI integration in enterprise-level life science research.