OpenAI launches GPT-Rosalind for life sciences

OpenAI frames GPT-Rosalind as the first release in a dedicated life sciences series, focusing on the complexity of research workflows—not just the science itself. The model's name honors Rosalind Franklin, whose work revealed DNA's structure, signalling an ambition to support foundational discovery. Key evaluations include outperforming GPT-5.4 on 6 of 11 LABBench2 tasks and strong results against human experts in a partnership with Dyno Therapeutics.

The launch combines a restricted model with a widely available plugin, aiming to serve both governed enterprise research and broader scientific use. The trusted-access model, with its eligibility and safety checks, suggests OpenAI is prioritizing controlled deployment over open availability. The inclusion of advisory partners like McKinsey and BCG points to a focus on enterprise integration.

The longer-term stakes hinge on whether these early evaluations translate into real, faster discoveries in drug development, where gains at the start compound downstream. As the model expands beyond the initial U.S. Enterprise preview, the test is whether it can maintain safety controls while scaling to a wider scientific community—and whether that produces genuinely better hypotheses than traditional methods. Partnerships with national labs like Los Alamos on protein design will offer early signals.