GPT-Rosalind pushes OpenAI into drug discovery with 50+ tools

OpenAI is moving deeper into scientific work with GPT-Rosalind, a purpose-built reasoning model for biology, drug discovery, and translational medicine. The model is not a broad consumer rollout: it is available as a research preview in ChatGPT, Codex, and the API for qualified customers through a trusted access program. That framing matters because the target workflow is not casual Q&A. OpenAI is aiming at evidence synthesis, hypothesis generation, experimental planning, data analysis, and tool-heavy research routines where a wrong shortcut can waste months in the lab.

The practical hook is the April 16 source post: OpenAI says scientists can use a Life Sciences research plugin for Codex that connects models to more than 50 public multi-omics databases, literature sources, biology tools, and repeatable skills. The plugin is freely accessible, while GPT-Rosalind itself starts with qualified Enterprise customers in the U.S. under eligibility, governance, and security review. OpenAI says research preview use will not consume existing credits or tokens, subject to abuse guardrails.

The benchmark claims are the reason this clears a higher bar than a normal vertical-AI launch. OpenAI says GPT-Rosalind achieved leading performance on BixBench among models with published scores, and beat GPT-5.4 on 6 of 11 LABBench2 tasks. The standout LABBench2 improvement was CloningQA, an end-to-end task for designing DNA and enzyme reagents for molecular cloning protocols. In a separate Dyno Therapeutics evaluation using unpublished RNA sequence data, best-of-ten model submissions ranked above the 95th percentile of 57 historical human expert scores on prediction and around the 84th percentile on sequence generation.

The early customer list also shows where OpenAI wants the model tested: Amgen, Moderna, Thermo Fisher Scientific, the Allen Institute, Novo Nordisk, NVIDIA, Benchling, UCSF School of Pharmacy, and Oracle Health and Life Sciences are named in the launch material. OpenAI also says it is working with Los Alamos National Laboratory on AI-guided protein and catalyst design.

The caveat is just as important as the headline. GPT-Rosalind is gated because biology models raise safety and misuse questions that normal productivity software does not. The near-term test will be whether a model that performs well on curated evaluations can reliably improve messy internal research workflows without increasing false confidence. If it does, drug discovery may become one of the first domains where frontier models are bought less as chatbots and more as governed research infrastructure.