OpenAI just dropped GPT-Rosalind, a reasoning model built for life sciences research and drug discovery. Named after Rosalind Franklin, whose work helped reveal DNA's structure, it targets the multi-step workflows that slow early-stage discovery: literature synthesis, experimental planning, hypothesis generation. Available now as a research preview in ChatGPT, Codex, and the API through a trusted access program. Partners include Amgen, Moderna, Novo Nordisk, Thermo Fisher, and Benchling.
The benchmark numbers look solid. GPT-Rosalind beat GPT-5.4 on 6 of 11 tasks in LABBench2, with the biggest jump in CloningQA, which tests end-to-end DNA design for molecular cloning protocols. In a partnership with Dyno Therapeutics on RNA sequence prediction using unpublished data, best-of-ten submissions ranked above the 95th percentile of human experts on prediction and around the 84th percentile on sequence generation. Strong numbers, though "best-of-ten" is generous framing.
DeepMind's AlphaFold solves specific structural biology problems with high precision. Isomorphic Labs, DeepMind's drug discovery spinout, uses those tools for therapeutic design. GPT-Rosalind takes a different angle: agentic reasoning and workflow orchestration rather than molecular physics. OpenAI is building a horizontal platform that slots into existing pharma ecosystems, with a free Life Sciences plugin for Codex connecting to over 50 scientific tools and databases. But real lab validation matters more than benchmarks.
The core pitch is straightforward. Drug discovery takes 10 to 15 years. Gains at the earliest stages compound downstream. Whether a language model can meaningfully compress that timeline is unproven, but the partner roster suggests organizations with real skin in the game are willing to find out.