OpenAI Unveils GPT-Rosalind for Science

OpenAI has introduced GPT‑Rosalind, a new frontier reasoning model engineered to accelerate scientific research in fields like biology, drug discovery, and translational medicine. This specialized model is optimized for complex scientific workflows, enhancing tool utilization and deepening comprehension across chemistry, protein engineering, and genomics.

The journey from identifying a drug target to regulatory approval can span over a decade. OpenAI believes advanced AI can significantly shorten this timeline by improving early-stage research, enabling scientists to explore more possibilities and formulate stronger hypotheses faster. GPT‑Rosalind is designed to assist with evidence synthesis, hypothesis generation, and experimental planning.

Built for Scientific Workflows

The GPT‑Rosalind life sciences model series is tailored for contemporary scientific endeavors, integrating published evidence, experimental data, and specialized tools. OpenAI's evaluations show superior performance in tasks involving reasoning over molecules, proteins, genes, and biological pathways. The model also demonstrates improved efficacy in multi-step workflows, including literature reviews, sequence-to-function interpretation, and experimental design.

This marks the initial release in the GPT‑Rosalind family, with plans to expand its biochemical reasoning capabilities for long-horizon, tool-intensive scientific tasks. OpenAI's robust compute infrastructure supports continuous training and refinement of these domain-specific models.

Ecosystem and Partnerships

OpenAI is collaborating with leading pharmaceutical, biotech, and research organizations, including Amgen, Moderna, and Thermo Fisher Scientific, to integrate GPT‑Rosalind into their discovery pipelines. These partnerships aim to leverage AI for accelerating AI for drug discovery and advancing AI for drug discovery.

A new Life Sciences research plugin for Codex is also available, connecting models to over 50 scientific tools and data sources. This plugin facilitates user interaction with human genetics, functional genomics, protein structure, and clinical evidence data.

Performance and Access

Evaluations indicate GPT‑Rosalind excels in core scientific reasoning tasks, outperforming previous models on benchmarks like BixBench and LABBench2. Notably, it demonstrated strong performance in DNA cloning protocol design and RNA sequence prediction and generation when tested in collaboration with Dyno Therapeutics, achieving top-tier rankings against human expert scores.

GPT‑Rosalind is launching via a trusted-access program for qualified enterprise customers in the U.S., with stringent safeguards against misuse. OpenAI is also making connectors and the Life Sciences Research Plugin more broadly available for use with their mainline models.

Organizations can request access through a qualification and safety review process. During this research preview, usage will not consume existing credits. OpenAI is committed to long-term development in AI for scientific discovery, with future work focusing on enhancing biological reasoning and expanding support for complex research workflows.