OpenAI Opens GPT-Rosalind to Research Labs — Outperforms GPT-5.5 on Biomedical Benchmarks
OpenAI expanded GPT-Rosalind on June 7 with two new life sciences plugins, giving qualified research organizations access to a biomedical AI that scores 63.2% on LabWorkBench versus GPT-5.5's 55.8%. Novo Nordisk is among the first users.
OpenAI expanded GPT-Rosalind on June 7, adding two domain-specific plugins that move the biomedical AI out of early preview and into controlled use by life sciences organizations.
The two additions — Life Sciences Research and Life Sciences NGS Analysis — give GPT-Rosalind sourced evidence retrieval, biomedical interpretation, and end-to-end bioinformatics execution. Researchers can now task the model with analyzing next-generation sequencing data, generating hypotheses from literature, and testing them against biomedical datasets without leaving the interface.
Performance numbers back the launch. GPT-Rosalind now scores 63.2% on LabWorkBench, the biomedical task benchmark, versus GPT-5.5’s 55.8% on the same suite — a meaningful lead in a domain where general-purpose models typically plateau. The model also uses 31% fewer tokens than GPT-5.5 on long-horizon quantitative biology tasks, which translates directly to lower API costs for organizations running large-scale analyses.
Novo Nordisk is publicly named as an early user. The Danish pharmaceutical giant is deploying GPT-Rosalind to analyze experimental datasets and pressure-test hypotheses before committing to wet-lab validation. That kind of use case — AI as a hypothesis filter upstream of expensive experiments — is exactly the efficiency lever drug discovery teams have been asking for.
Access remains tightly gated. OpenAI operates a trusted-access program restricting GPT-Rosalind to qualified research organizations: academic institutions, hospitals, pharmaceutical companies, and biotech firms. There is no general consumer or API access as of June 8. OpenAI says the phased rollout is intentional given the sensitivity of healthcare data and the need to validate model outputs against expert review.
GPT-Rosalind first shipped in April 2026, focused initially on genomic data interpretation. The June 7 plugin expansion is its first major capability update, signaling OpenAI’s intent to build a dedicated vertical AI product line for life sciences — separate from ChatGPT’s generalist track.
The LabWorkBench benchmark tests biomedical reasoning across clinical trial design, drug-target interaction, and lab protocol generation. A 7.4-point lead over GPT-5.5 on that suite — at lower token cost — makes GPT-Rosalind the strongest generative AI tool in the sector by these metrics.
Whether the restricted-access model scales into commercial deployment is the open question. Pharmaceutical companies pay substantial software license fees for specialized tools. An OpenAI model that demonstrably accelerates drug discovery could command pricing far above standard API rates.