Anthropic launches Claude Science, an AI workbench built for the lab

Anthropic has pushed decisively past the chatbot. On 30 June 2026 it launched Claude Science, an AI 'workbench for scientists' that folds more than 60 scientific databases, computing tools and research workflows into a single environment. The intended user is the life-sciences researcher whose day is currently splintered across PubMed, Jupyter notebooks, R and a cluster terminal; Claude Science's proposition is to collapse that stack into one surface where the model can read the literature, run the analysis and manage the workflow without the researcher switching tools. It is in beta for Claude Pro, Max, Team and Enterprise on macOS and Linux.

The important distinction is that this is not merely 'Claude, but told to think about biology'. It is a product built around integration — connective tissue between the model and the fragmented apparatus of real research. That framing matters because it changes where the competitive advantage sits. Raw model intelligence is increasingly a commodity that every frontier lab can approximate; what a working scientist cannot easily assemble is a single environment that already speaks to dozens of curated databases, holds computing tools to hand, and preserves the workflow across a long project. Anthropic is betting the durable value is in that plumbing, not in a benchmark score.

Claude Science did not appear from nowhere. It is the visible edge of a deliberate campaign. In June 2026 Anthropic acquired the computational-biology startup Coefficient Bio in an all-stock deal worth roughly $400 million, and it hired John Jumper — the researcher who led AlphaFold at Google DeepMind and shared the 2024 Nobel Prize in Chemistry for it. Both moves feed a single strategic thesis that CEO Dario Amodei has stated openly: to compress life-sciences R&D cycles by a factor of 10x. A workbench is how that thesis becomes a product a researcher can actually open on a Monday morning.

AI-for-science is now an explicit frontier-lab battleground. Anthropic is racing OpenAI, which shipped GPT-Rosalind — a model tuned for biological reasoning — in April 2026, and Google DeepMind, the institution Jumper left. But the more revealing detail is on the customer side. Reporting notes that Novo Nordisk is a named partner for both Anthropic and OpenAI, and that the Allen Institute has worked with both as well. That overlap is a tell: big pharma is running parallel evaluations, not anointing a single vendor. The land grab is real, but as of now it is unwon — no lab has converted early access into lock-in.

To seed adoption, Anthropic paired the launch with a grants programme: up to 50 projects, each eligible for as much as $30,000 in Claude credits, with the compute platform Modal adding up to $2,000 of compute for select projects. Applications are open through 15 July 2026, awards land by 31 July, and funded projects run 1 September to 1 December 2026. The commercial logic is straightforward — subsidised credits get the workbench into real labs, and every funded project generates the workflow patterns and usage data that make the product stickier. The moat Anthropic is digging is not model IQ; it is workflow, data integration and scientific credibility, and grants are the cheapest way to start filling it.

The honest read as of early July is that Anthropic has assembled an unusually coherent bet — talent, an acquisition, a workbench and a funding on-ramp all aimed at the same goal — but that the outcome hinges on adoption it does not yet control. If researchers make Claude Science the place they actually work, the data and switching costs compound in Anthropic's favour. If pharma keeps evaluating three vendors in parallel, the workbench is a strong opening move in a game that is only beginning.