- Anthropic released Claude Science on Tuesday, a single workspace for scientific research.
- Anthropic is already using it for drug discovery work on neglected tropical diseases.
- Google and OpenAI have both launched rival science tools within the past two months.
Anthropic released Claude Science, an application built to give scientists and pharmaceutical researchers a single workspace for their work, rather than a new AI model.
It arrives just weeks after Google and OpenAI made near-identical moves, turning what was a niche corner of the AI market into a three-way fight over who gets to sit next to scientists at the lab bench.
A single workspace instead of a dozen tools
Anthropic built Claude Science to solve a specific complaint from scientists: they spend their days switching between databases, coding notebooks, terminals and viewers instead of doing research.
The tool integrates the packages researchers already use, produces auditable outputs and gives flexible access to computing power, running locally on macOS or Linux, or on a remote machine over SSH or a high-performance computing login node.
Every result carries a trail back to the code and environment that produced it, along with a plain-language explanation and the full message history behind it, aimed at making reproducibility less of a headache for scientists. Users interact with a coordinating agent that has access to more than 60 databases and skills, and can call in specialist agents as needed.
The company is drawing on NVIDIA’s BioNeMo Agent Toolkit to plug in specialised life-science models including Evo 2, Boltz-2 and OpenFold3, while heavier computational jobs like protein folding or genomics pipelines route out to a lab’s existing HPC clusters or a Modal account. Because Claude Science runs on Anthropic’s existing Claude models, it has already been through the company’s standard safety and biosecurity evaluations, rather than requiring a new review process.
Claude Science is available now in beta to Claude Pro, Max, Team, and Enterprise users on macOS and Linux, with admins needing to switch it on for Team and Enterprise plans, and Anthropic offering discounted Team plans to research labs. Anthropic will also back up to 50 AI for Science projects with as much as $30,000 in credits each, with applications open through July 15, 2026, and award notifications going out by July 31.
A race that started months before launch
Anthropic is not first to this idea, and it is not alone in it, either.
Google launched its own “Gemini for Science” workbench at its I/O conference in May, pitched the same way, a scientific workbench bundling skills across more than 30 life-science databases. OpenAI moved even earlier, releasing GPT-Rosalind in April, an AI model aimed squarely at speeding up research and drug discovery.
That timing is not a coincidence. All three companies are chasing the same prize: locking in scientists and pharmaceutical researchers as long-term enterprise customers, at a moment when Anthropic and OpenAI are both racing toward IPOs and need new revenue lines to justify their valuations. Anthropic confidentially filed its S-1 with the United States Securities and Exchange Commission in June, four days after closing a $65 billion round at a $965 billion valuation, and reports suggest a listing as early as October.
The line worth reading twice
Underneath the competitive framing sits a detail that has nothing to do with market share. Anthropic is already using Claude Science internally for its own pre-clinical drug discovery work, including on neglected tropical diseases that rarely attract commercial investment because there is little profit in curing them. Northeastern researcher Michael Pollastri, whose work focuses on repurposing existing drugs for tropical diseases, said identifying candidates is normally a slow process, and that Claude Science automating the information-gathering side could genuinely speed that up.
It is not the line Anthropic led with, and it will not be the one Google or OpenAI can easily copy. Tools can be matched. A pipeline built around diseases nobody else will fund is harder to replicate on a press release timeline.
