OpenAI just gave GPT-Rosalind, its specialized model for life sciences research, a major capability upgrade. According to OpenAI, the refreshed model pushes harder into the parts of biology research that actually move drug programs forward: deeper biological reasoning, medicinal chemistry expertise, genomics analysis, and support for full experimental workflows. GPT-Rosalind, named after DNA pioneer Rosalind Franklin, first launched as a research preview earlier this year. This update is about making it sharper where scientists need it most.
What stands out here is the focus. OpenAI isn’t pitching a general assistant that happens to know some biology. It’s building a model tuned for the messy, multi-step reality of lab work.
What’s new in this update
Here’s what OpenAI says the upgraded GPT-Rosalind brings to the bench:
- Stronger medicinal chemistry reasoning. The model handles complex chemistry queries with better multimodal synthesis and mechanistic reasoning, the kind of work that sits at the center of designing a drug candidate.
- Better genomics analysis. OpenAI reports gains on genomics interpretation and pathway analysis, areas where researchers drown in data and need help connecting signals to mechanisms.
- Experimental workflow support. GPT-Rosalind now does more across multi-step research tasks, including quantitative biology and wet lab troubleshooting, not just answering one-off questions.
- More efficient performance. On the GeneBench evaluation, OpenAI says GPT-Rosalind hits 21.6% accuracy versus 20.4% for GPT-5.5, while using 31% fewer tokens. Higher accuracy at lower cost is the combination that matters when you’re running thousands of queries.
How it compares
GPT-Rosalind builds on GPT-5.5’s agentic coding and tool-use skills, then layers in deeper domain intelligence for drug discovery. So you get the general model’s ability to chain tools and write code, plus specialist knowledge in chemistry and genomics. That token efficiency gain over GPT-5.5 is the clearest signal that this isn’t just GPT-5.5 with a biology label. It’s a model trained to do this specific job better and cheaper.
Who can use it, and when
This is where the launch stays deliberately narrow. GPT-Rosalind is available now as a research preview, but only to eligible organizations through OpenAI’s trusted-access program. You’ll find it in ChatGPT, Codex, and the API for institutions accepted into that program.
This isn’t a model anyone can sign up for today. OpenAI is gating access to vetted research and pharma groups. Early partners include Amgen, Moderna, Thermo Fisher Scientific, and the Allen Institute, all working with OpenAI to apply the model across R&D workflows.
Why it matters
Drug discovery is slow and brutally expensive. A single candidate can take years and billions of dollars to move from idea to clinic, and most fail along the way. Any tool that shaves time off target discovery, validation, or literature synthesis has real leverage on that math.
That’s the bet OpenAI is making. Instead of a model that’s broadly smart, it’s offering one that speaks the language of the lab and plugs into the actual tools scientists use. OpenAI has paired GPT-Rosalind with a Life Sciences research plugin for Codex that connects models to more than 50 scientific tools and data sources, which tells you the company wants this thing inside real research pipelines, not sitting in a chat window.
The caveats
Keep the limits in view. This is a research preview, not a finished product, and access is restricted to approved institutions. The benchmark gains, while real, are incremental rather than dramatic, a couple of percentage points on GeneBench. And as with any AI used in science, the model’s output still needs expert validation. No serious lab is going to take a hypothesis and run with it unchecked.
The bigger story is direction. OpenAI is clearly building a family of specialized scientific models, and GPT-Rosalind is the flagship for biology. If the trusted-access approach works, expect the partner list to grow and the capabilities to keep compounding. For now, watch how Amgen, Moderna, and the research institutes put it to work. Their results will tell us whether domain-specific AI can really speed up discovery, or just talk a good game. Full details are available at the original source.