GPT-Rosalind Update: Enhanced Life Sciences Capabilities and Research Tools

OpenAI has released a model update to the GPT-Rosalind series, a model specifically engineered for enterprise-scale life sciences research. This update integrates the agentic coding and tool-use capabilities of GPT-5.5 with enhanced intelligence in core drug-discovery domains, including genomics and medicinal chemistry, to improve performance across analysis, design, and experimental workflows.

Performance Gains in Life Sciences Research

GPT-Rosalind demonstrates broad performance improvements on research tasks involving quantitative biology, complex medicinal chemistry queries, and wet lab troubleshooting. To measure these gains, OpenAI developed LifeSciBench, an externally expert-judged benchmark that evaluates end-to-end scientific workflows across six key areas:

• Evidence handling
• Analysis
• Design and optimization
• Scientific reasoning
• Validation and operations
• Translation and communication

Medicinal Chemistry

Using the MedChemBench evaluation—which covers multimodal chemical structure understanding, structure-activity relationship (SAR), ADME prediction (absorption, distribution, metabolism, excretion), lead-optimization, and retrosynthesis—GPT-Rosalind outperformed GPT-5.5 with a score of 27.5% compared to 25.1%, while utilizing 7.2% fewer tokens.

Genomics and Quantitative Biology

On GeneBench, an agentic evaluation focusing on long-horizon, end-to-end analysis in proteomics, epigenomics, spatial transcriptomics, and functional genomics, GPT-Rosalind achieved 21.6% accuracy (compared to 20.4% for GPT-5.5) while using 31% fewer tokens.

Wet Lab Assistance

OpenAI introduced LabWorkBench to test the model's ability to link perturbations to experimental outcomes in real wet lab protocols. GPT-Rosalind scored 63.2%, significantly higher than GPT-5.5's 55.8%, while reducing token usage by 5.3%.

Integrated Research Workflows and Tooling

To transition from model reasoning to executed workflows, OpenAI has released two plugins accessible through Codex:

• Life Sciences Research Plugin: Provides sourced evidence retrieval and biological interpretation.
• Life Sciences NGS Analysis Plugin: Enables bioinformatics execution, such as turning processed ctDNA records into interactive notebooks or converting 10x-style matrix bundles into QC-filtered single-cell artifacts and UMAPs.

These plugins are supported by new interactive viewers for biologically native file types, including sequence, alignment, and structure viewers, allowing researchers to inspect evidence (such as mutant residues or inhibitor-bound pockets) directly within the workspace.

Deployment and Enterprise Access

GPT-Rosalind is available in research preview to eligible organizations globally via a trusted-access deployment structure. This access is restricted to organizations conducting legitimate scientific research with clear public benefit, strong governance, and enterprise-grade security.

As part of this rollout, OpenAI is partnering with Novo Nordisk to scale medical research, utilizing GPT-Rosalind to connect evidence across literature, genomics, transcriptomics, and experimental results to accelerate drug discovery.

"Life sciences research is complex, data-rich, and interdisciplinary. To deliver meaningful value for researchers, advanced AI models must be grounded in trusted scientific data, connected to validated tools, and integrated into the real-world workflows researchers use every day." — Mishal Patel, Group Vice President, AI & Digital Innovation, R&D - Novo Nordisk

Future Directions

OpenAI intends to continue improving GPT-Rosalind's biological reasoning and support for long-horizon research workflows. The model's capabilities are being applied to high-impact public-benefit work, including translational medicine, public health, and biodefense via the Rosalind Biodefense initiative.