OpenAI Boosts Life-Sciences AI Model GPT-Rosalind
OpenAI announced on June 3, 2026, that it is bringing new capabilities to GPT-Rosalind, a model series purpose-built for life sciences research at enterprise scale. The update merges the agentic coding and tool use of its latest GPT-5.5 with stronger intelligence for drug discovery, analysis, design, and experimental workflows.
The announcement came via the company's official X account (@OpenAI). GPT-Rosalind was first unveiled on April 16, 2026, as "Introducing GPT‑Rosalind for life sciences research," and this represents an update to that model. By incorporating GPT-5.5's autonomous coding and tool use, it lifts performance as a frontier reasoning model for biology, drug discovery, and translational medicine. The name honors Rosalind Franklin, the scientist who helped reveal the structure of DNA. (OpenAI)
In the pharmaceutical industry, it takes an average of 10 to 15 years from target discovery to approval, making efficiency in hypothesis generation, evidence integration, and experimental planning at the early discovery stage a major challenge. GPT-Rosalind supports multi-step workflows that span literature, databases, and tools, aiming to accelerate researchers' work. For OpenAI, the notable point is that it squarely positions a domain-specific (life sciences) model rather than a general-purpose one; unlike structure-biology-focused systems such as AlphaFold, it serves as a reasoning and orchestration layer. (Fierce Biotech)
It is offered as a research preview, limited to vetted U.S. enterprise customers. It is available in ChatGPT Enterprise, Codex, and the API, with API use restricted to internal research tools and workflows—not customer-facing products or external commercial apps. Access is reviewed through an application form, and during the preview it is free (with usage limits), consuming no existing credits or tokens. For Codex, a "Life Sciences research plugin" connecting more than 50 scientific tools and databases is published on GitHub, supporting protein structure lookup, sequence search, and literature review. On security, it features SOC 2 Type 2, HIPAA-aligned controls, and role-based access control, and OpenAI states it does not use customer data for training. (Help Center)
On performance, it showed leading results on BixBench (bioinformatics/data analysis), and on the research-workflow benchmark LABBench2 it surpassed the prior GPT-5.4 on 6 of 11 tasks, with notable gains on CloningQA, which tests DNA/enzyme reagent design. In a private RNA sequence-to-function task with Dyno Therapeutics, it reportedly scored above the 95th percentile on a prediction task and at the 84th percentile on sequence generation, compared with 57 historical scores from human experts. (OpenAI)
Reaction on X from insiders was largely positive. OpenAI's Greg Brockman (@gdb) called it a "Major upgrade" to GPT-Rosalind, highlighting better intelligence for drug discovery. Life Sciences lead Chris Hayduk (@ChrisHayduk) praised the integration of advances in agentic coding, biological/chemical reasoning, and Codex tooling, as well as its value as a dynamic workbench that keeps scientists in the loop via interactive viewers. Jeremy (@jeremyli__), in lifesci consulting, cited concrete improvements such as token efficiency and performance on GeneBench. Some general users analyzed it as a vertical model showing that domain data and lab feedback loops will beat raw scale, while others voiced skepticism about the limited nature of trusted access. (Tahseen_Rahman)
Separately, OpenAI is also advancing a "Rosalind Biodefense" program that extends GPT-Rosalind to public health and biosecurity, framing it as strengthening societal resilience. (Rosalind Biodefense)