OpenAI unveils GPT-Rosalind, a model to speed up drug discovery and genomics

OpenAI has unveiled GPT-Rosalind, a specialized frontier reasoning model designed to accelerate work in the life sciences. The company positions it as a new class of tool, able to handle biomedical tasks at a level general-purpose assistants cannot reach. The model targets four key areas: discovering new drugs, analyzing genomic data, reasoning about proteins and supporting scientific research workflows.

The name is a deliberate nod to Rosalind Franklin, the British crystallographer whose X-ray images of DNA's structure in the early 1950s proved key to understanding the double helix. Although Franklin never received proper recognition during her lifetime, her contribution was critical. The choice of name is symbolic: OpenAI stresses its ambition to build a tool that could become a turning point for modern biology, much as her images changed molecular science in the last century. Only now, the role of "Photograph 51" is played by AI.

GPT-Rosalind is built on OpenAI's frontier extended-reasoning architecture and adapted to the specifics of life sciences tasks. Unlike general-purpose models such as GPT-4o or o3, it is optimized for scientific data that demands precise biomedical interpretation. Key applications include analyzing genomic sequences, predicting protein functions and interactions, forming hypotheses about the mechanisms of action of drug compounds, and synthesizing results from clinical studies and patent databases. In other words, the model's job is to reason like an experienced biochemist, not like a search engine.

The market for AI-assisted drug development is in a phase of active growth. By industry estimates, bringing a single drug to market takes 10 to 15 years and costs billions of dollars in investment. Most of that time and budget goes to the early stages: selecting candidate molecules, preclinical studies and toxicity prediction. This is where AI models show the greatest potential to cut time and cost.

Google DeepMind's AlphaFold has already changed structural biology by offering a working solution to the problem of predicting the three-dimensional structure of proteins, a problem scientists could not formalize for decades. GPT-Rosalind enters this market from a different angle: not structural computation, but reasoning over heterogeneous scientific data and the ability to build well-founded hypotheses.

The model's key task is not retrieving information but constructing reasoned scientific hypotheses from diverse sources: scientific publications, clinical protocols, genomic databases and patent registries. This requires multi-level reasoning and the ability to hold complex scientific context, which is exactly what sets the latest generation of frontier models apart from traditional search engines and narrowly specialized bioinformatics tools.

OpenAI is aiming GPT-Rosalind at three key segments: biopharmaceutical companies searching for new drug compounds; genomic and medical research institutes working with large-scale biological data; and academic laboratories conducting fundamental research. The plan is integration through an API into existing scientific pipelines, without the need to overhaul workflows entirely.

The launch of GPT-Rosalind is part of OpenAI's broader verticalization strategy. After the general-purpose GPT-4o and the o-series, the company is steadily moving toward domain-specific models for particular professional tasks. Competitors are heading the same way: Google DeepMind is developing Med-Gemini and AlphaFold 3, Microsoft is building AI infrastructure for clinical systems, and biotech startups such as Recursion Pharmaceuticals and Insilico Medicine are building AI-first pipelines to develop drugs from scratch.

Specialized AI models for science are moving from experimental tools into production infrastructure. For research teams working with genomic data or running preclinical trials, GPT-Rosalind is an additional tool in the stack that strengthens scientific expertise rather than replacing it. How much it will actually speed up real discoveries will be shown by its use in large research organizations.