Novo Nordisk, the Danish diabetes and obesity specialist, has just sealed an agreement with Anthropic.
The deal goes beyond pure R&D. It also touches internal operations modernization, up to “agentic” software development alongside drug discovery itself.
“AI tools can also unlock entirely new scientific opportunities and help illuminate human biology and how drugs work,” explains Mike Doustdar, the CEO of Novo Nordisk.
This closer collaboration adds to Novo Nordisk’s existing arrangements with OpenAI and Amazon Web Services, as well as Anthropic’s growing roster of pharmaceutical clients, which had already given Claude access to Bristol Myers Squibb teams for their drug research and development work.
Between 2025 and 2026, the world’s leading laboratories – Eli Lilly, Novartis, Johnson & Johnson, Novo Nordisk, Merck – have thus multiplied alliances with AI giants, whether model providers (OpenAI, Anthropic, Google/DeepMind), infrastructure (Nvidia) or cloud (Google Cloud, Microsoft), with commitments potentially exceeding one billion dollars per contract.
But Novo Nordisk plans to first test Anthropic’s models and its Claude Science environment before contemplating broader deployment.
Claude Science, a general-purpose scientificWorkbench
Launched in late June in public beta, Claude Science is a workbench that builds on Claude models already included in the user’s subscription, adding scientific tools, database connections, and computational integrations.
Concretely, the tool connects to more than sixty scientific databases and carries prebuilt toolkits for genomics, proteomics, structural biology, or computational chemistry. Its architecture rests on a main agent able to call on specialized sub-agents to distribute tasks, with a dedicated proofreading agent responsible for verifying citations and calculations before any publication.
Early users, notably at the Allen Institute (Seattle) and the UCSF Brain Tumor Center (San Francisco), reported significant reductions in the time needed to run complex analyses.
OpenAI counters with GPT-Rosalind
Facing Anthropic, OpenAI unveiled in April 2026 a model tailor-made for life sciences.
GPT-Rosalind is a reasoning model designed specifically to support scientific research in biology, drug discovery, and translational medicine. It is optimized for complex workflows that blend advanced reasoning with deep integration with tools, databases, and the sector’s experimental processes.
Partnered with heavyweight players, OpenAI can already rely on Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific as pilot partners.
Two strategies, one aim
There is, however, a marked difference between their two approaches.
Anthropic bets on a transversal scientific reasoning backbone, deeply integrated into laboratories’ internal workflows through contracts, such as with Bristol Myers Squibb covering the entire R&D and operations chain.
OpenAI, on its side, has chosen full vertical integration with a dedicated model designed from the outset for biology and backed by the broader ChatGPT ecosystem, the API, and Codex.
Nothing suggests these two logics are mutually exclusive. The same laboratory can use Claude Science for transversal scientific reasoning tasks while relying on GPT-Rosalind for precise genomics or molecular design work. Novo Nordisk provides a good example of this coexistence.
Becoming the infrastructure… not just the license
Beyond the announcements, the battle centers on the software infrastructure of pharmaceutical R&D; a role traditionally occupied by LIMS (Laboratory Information Management Systems), electronic laboratory notebooks, or high-performance computing platforms.
For labs, the spread of these partnerships unlocks three levers: balancing between a generalist and a vertical approach (or blending both), seeking measurable productivity gains across discovery cycles that have historically been lengthy, and pitting vendors against each other to negotiate access terms and governance safeguards for data. Novo Nordisk has underscored this by citing human oversight protocols aligned with its compliance requirements.
——————————————————————————————————–
Claude Science (Anthropic) vs Rosalind Workbench (OpenAI) — a comparison
| Dimension | Claude Science (Anthropic) | Rosalind Workbench (OpenAI) |
|---|---|---|
| Type of offering | Integrated workbench within the Claude ecosystem, designed for research workflows and scientific reasoning. | Dedicated scientific workspace for the life sciences, integrated with the ChatGPT application. |
| Main target | R&D and scientific research teams across all domains (biopharma, chemistry, physics, etc.), with a strong emphasis on biopharma. | R&D teams in life sciences (biopharma, biotech, biology research institutes). |
| Underlying model | Anthropic frontier models (Claude), optimized for scientific reasoning and tool usage. | Specialized GPT-Rosalind model, pitched as OpenAI’s top performer for life sciences research. |
| Key features | – Advanced scientific reasoning – Orchestration of multi-step workflows – Data analysis, code generation, report writing – Integration into R&D pipelines and agentic software. |
– Guided workflows for biology (structures, sequences, genomics, medicinal chemistry) – Integrated bioinformatics tools (3D visualization, alignment, NGS QC, etc.) – Research plugin connected to over 50 tools and databases – Explore mode (general questions) and Research mode (multi-step workflows). |
| Use cases announced | – Target discovery and molecule design – Genomic/proteomic data analysis – Writings and synthesis of clinical study reports – Protocol and R&D workflow optimization – In-house “agentic” software development. |
– Protein structure and sequence analysis – Genomics and NGS analyses (FASTQ QC, RNA-seq bulk/single-cell) – Medicinal chemistry and small-molecule design – Experiment planning and literature reviews. |
| Access mode | – Enterprise access via API and integrations – Deployment on targeted workflows, with data governance and human supervision. |
– Enterprise offering via ChatGPT Enterprise, Codex (with plugin) and API – Access “trusted access” for the Research mode, restricted to verified organizations. |
| First announced partners / users | – Bristol Myers Squibb (enterprise deployment) – Novo Nordisk (pilot on targeted R&D workflows) – Research institutions (Allen Institute, UCSF Brain Tumor Center) cited as early users. |
– Amgen – Moderna – Allen Institute – Other drug-discovery teams with controlled access. |
| Strategic positioning | Generalist scientific workbench, heavily oriented toward biopharma, integrated into a broader strategy of automating science and enterprise deployment of Claude. | Life sciences–focused workbench, presented as OpenAI’s flagship offering for R&D in biology and drug discovery. |
| Governance / security | Data governance measures and human oversight, aligned with clients’ compliance standards (e.g., Novo Nordisk). | Controlled access (enterprise programs and trusted access), with emphasis on traceability of results and human review of Research workflows. |
