# OpenAI introduces GPT-Rosalind for life-sciences research > The purpose-built reasoning model enters trusted research preview with a Codex plugin connecting more than 50 scientific tools and data sources. Canonical URL: https://www.devobs.io/news/news-openai-gpt-rosalind-life-sciences-preview/ By: Owen Park Published: 2026-09-06T11:58:54.623Z Updated: 2026-09-06T11:58:54.623Z Event date: 2026-04-16 Section: AI OpenAI introduced [GPT-Rosalind](https://openai.com/index/introducing-gpt-rosalind/) on April 16 as a purpose-built reasoning model for biology, drug discovery, and translational medicine. The model entered research preview in ChatGPT, Codex, and the API for qualified customers through a trusted-access program. OpenAI also released a Life Sciences research plugin for Codex that connects workflows to more than 50 public scientific tools and data sources. ## The product targets research workflows The release focuses on evidence synthesis, sequence and protein analysis, hypothesis generation, experimental planning, and tool use across multi-step work. OpenAI evaluated the model on organic chemistry, protein understanding, genomics, experimental design, and scientific tool selection. It also reported results on BixBench, LABBench2, and an uncontaminated RNA sequence task supplied by Dyno Therapeutics. Those are vendor-reported evaluations and do not establish performance for a laboratory’s specific data or decision process. The Codex plugin is independently useful as an integration layer. Its modular skills cover areas such as human genetics, functional genomics, protein structure, biochemistry, clinical evidence, and study discovery. Researchers can inspect the chosen sources and tools rather than treating the model response as the final scientific record. ## Access control is part of the deployment OpenAI limits GPT-Rosalind to organizations with beneficial research uses, governance and safety oversight, and controlled enterprise access. That boundary matters because stronger biological assistance can be dual-use and because research data may be sensitive. Teams evaluating the preview should test provenance, reproducibility, unsupported-citation rates, tool permissions, dataset retention, and expert sign-off. A model can accelerate exploration without becoming the authority for experimental validity. Preserve inputs, tool versions, intermediate artifacts, and human decisions so a result can be audited and repeated outside the chat session. ## Source references -