Anthropic on June 30 launched Claude Science in beta, an AI workbench purpose-built for scientific research that integrates more than 60 pre-configured databases and toolkits covering genomics, proteomics, structural biology, and cheminformatics.
Unlike Claude Code or Claude Cowork, Claude Science does not introduce a new underlying model. It runs on existing Claude models — including Opus 4.8 — and focuses on workflow rather than raw capability, consolidating tools that researchers typically navigate across dozens of separate platforms.
How it works
At its core, a central AI assistant acts as a project manager, capable of spawning specialist sub-agents to carry out distinct tasks in parallel. A built-in reviewer agent independently checks citations and flags computational inconsistencies. All outputs — figures, manuscripts, 3D protein structures, and genome tracks — are accompanied by auditable code histories and full message logs, addressing the reproducibility demands of peer-reviewed science.
Compute management is handled automatically: workloads can run on a researcher’s local machine, institutional HPC clusters, or on-demand GPUs through a partnership with Modal, without manual setup. The platform also integrates with NVIDIA’s BioNeMo toolkit and can connect proprietary datasets as reusable skills, keeping sensitive data on existing institutional infrastructure rather than Anthropic’s servers.
Access and pricing
Claude Science is available in beta to Claude Pro, Max, Team, and Enterprise subscribers. Academic institutions and nonprofit research organizations qualify for a discounted Team plan.
To accelerate early adoption, Anthropic is funding up to 50 AI for Science projects, each receiving up to $30,000 in Claude credits and up to $2,000 in Modal compute grants. Applications are open through July 15, 2026, with projects running September through December. Postdoctoral researchers and graduate students working in biology and biomedical science are the primary audience.
A three-way race in scientific AI
Claude Science enters a market where three distinct strategies are now competing: Anthropic’s broadly accessible subscription model, OpenAI’s more narrowly gated GPT-Rosalind (launched in April 2026 for enterprise customers), and Google DeepMind’s approach of bundling proprietary models such as AlphaFold and AlphaGenome alongside 30+ databases within Gemini for Science.
By placing Claude Science alongside Claude Code as a flagship product — rather than a specialized add-on — Anthropic has signaled that scientific computing is a long-term strategic priority, not an experiment.
An early case study from the Allen Institute found that neuroscientist Jérôme Lecoq was able to dramatically compress the time required for a major literature review using Claude Science’s multi-agent system, according to Anthropic.