OpenAI introduced an updated version of GPT-Rosalind on June 3, 2026, targeting enterprise-scale life sciences research. The model integrates GPT-5.5's agentic coding and tool-use capabilities with enhanced domain intelligence in medicinal chemistry and genomics. It is currently accessible to eligible global organizations through a research preview that requires strict governance, enterprise-grade security, and a focus on public benefit. The company also announced a collaboration with Novo Nordisk to scale medical research workflows.

To measure real-world performance, OpenAI developed LifeSciBench, a benchmark encompassing six key areas: evidence handling, analysis, design, scientific reasoning, validation, and communication. The updated model demonstrated significant gains over GPT-5.5 across specialized tasks. In medicinal chemistry, GPT-Rosalind achieved a 27.5% score on MedChemBench compared to 25.1%, while reducing token usage by 7.2%.

On GeneBench, the model reached 21.6% accuracy against 20.4% while using 31% fewer tokens. Testing on LabWorkBench, which evaluates wet lab protocol troubleshooting, resulted in a score of 63.2% compared to 55.8% for GPT-5.5, with 5.3% fewer tokens. To support practical execution, OpenAI introduced the Life Sciences Research and Life Sciences NGS Analysis plugins, enabling evidence retrieval and bioinformatics execution within a single workspace.

OpenAI introduced an updated version of GPT-Rosalind on June 3, 2026, targeting enterprise-scale life sciences research. The model integrates GPT-5.5's agentic coding and tool-use capabilities with enhanced domain intelligence in medicinal chemistry and genomics. It is currently accessible to eligible global organizations through a research preview that requires strict governance, enterprise-grade security, and a focus on public benefit. The company also announced a collaboration with Novo Nordisk to scale medical research workflows.

To measure real-world performance, OpenAI developed LifeSciBench, a benchmark encompassing six key areas: evidence handling, analysis, design, scientific reasoning, validation, and communication. The updated model demonstrated significant gains over GPT-5.5 across specialized tasks. In medicinal chemistry, GPT-Rosalind achieved a 27.5% score on MedChemBench compared to 25.1%, while reducing token usage by 7.2%.

On GeneBench, the model reached 21.6% accuracy against 20.4% while using 31% fewer tokens. Testing on LabWorkBench, which evaluates wet lab protocol troubleshooting, resulted in a score of 63.2% compared to 55.8% for GPT-5.5, with 5.3% fewer tokens. To support practical execution, OpenAI introduced the Life Sciences Research and Life Sciences NGS Analysis plugins, enabling evidence retrieval and bioinformatics execution within a single workspace.