TL;DR:

• OpenAI has released GPT-Rosalind, a frontier model tuned for biology, protein engineering and genomics, via a gated “trusted access” programme that limits initial use to qualifying US enterprises.
• The launch includes a free Life Sciences research plugin for Codex that connects models to over 50 public multi-omics databases and biology tools.
• The model beat human experts at RNA sequence-to-function tasks in benchmarks co-designed with Dyno Therapeutics — a pointed contrast with the UK government’s same-day Sovereign AI bet on domestic drug-discovery startups.

Named after Rosalind Franklin, GPT-Rosalind is OpenAI’s first model in a new life sciences series. It is available as a research preview in ChatGPT, Codex and the API for qualified customers, with launch partners including Amgen, Moderna, Thermo Fisher Scientific, Oracle Health and Life Sciences, the Allen Institute, NVIDIA, Benchling and UCSF School of Pharmacy.

What the model does

OpenAI positions GPT-Rosalind as a reasoning model optimised for scientific workflows — literature review, sequence-to-function interpretation, experimental planning and multi-step data analysis. On LABBench2, a benchmark covering research tasks from literature retrieval to protocol design, it outperformed GPT-5.4 on six of 11 tasks, with CloningQA — end-to-end design of DNA and enzyme reagents — showing the largest gain. In a private evaluation co-run with Dyno Therapeutics on uncontaminated RNA sequences, best-of-ten submissions scored above the 95th percentile of human experts on prediction tasks.

Access and controls

Trusted access sits on three tests: beneficial use, governance with misuse-prevention, and enterprise-grade security. Customers must onboard through OpenAI’s qualification and safety review, and US enterprises are first. The plugin for Codex is unrestricted and usable with mainline models — a practical route for UK research teams whose institutions cannot meet the enterprise prerequisites today. OpenAI says pricing during the preview will not consume existing credits, subject to abuse guardrails.

UK context

The release lands the same day the UK unveiled its first Sovereign AI equity stakes, with biological-foundation-model startup Prima Mente and RNA therapeutics firm Twig Bio among the recipients of government-backed compute. The contrast is pointed: OpenAI and US pharma giants have a gated, vertically integrated AI discovery stack; the UK is betting on distributed startups that will likely consume APIs like this one. For UK pharma R&D leads, GPT-Rosalind lowers the barrier to AI-assisted triage of literature and sequences — but the trusted-access gating means strategic workloads may still need to sit inside the OpenAI enterprise perimeter.

Looking forward

Expect a fast-follow from Anthropic and Google DeepMind. Isomorphic Labs, DeepMind’s London-based drug-discovery arm, already targets this space and will have to respond. The next test of GPT-Rosalind is not benchmarks but citation rate: how often does a peer-reviewed paper attribute a meaningful design decision to the model?