Claude Science: Anthropic's AI Workbench for Scientists

Scientists today juggle PubMed, Jupyter notebooks, R scripts, cluster terminals, and domain-specific viewers — often losing time and reproducibility in the shuffle. On June 30, 2026, Anthropic released Claude Science, a desktop AI workbench that brings these fragmented tools together into a single, auditable research environment . Announced at Anthropic's invite-only "The Briefing: AI for Science" event, Claude Science is designed to accelerate research by letting scientists analyze literature, run multi-step analyses, create figures and manuscripts, and trace every output back to its source code and environment .

Claude Science is not a new AI model — it runs on the same Claude models users already access (Opus 4.8 in its initial release) — but rather a purpose-built app that layers scientific tooling, database connectors, compute orchestration, and automated quality control on top of Claude's reasoning capabilities .

Key Features

Unified research environment — Claude Science replaces the workflow of jumping between PubMed, Benchling, Jupyter, R, cluster terminals, and chemical drawing software by integrating them all into one workspace . Researchers can conduct literature reviews, execute code, generate figures, and iterate on manuscripts without leaving the app .

Rich, reproducible artifacts with auditable history — Every figure, table, or manuscript is generated alongside the exact code and environment that produced it, plus a full message history. This makes validation and reproduction straightforward and directly addresses the reproducibility crisis in computational research .

Native scientific visualization — Claude Science natively renders 3D protein structures, genome browser tracks, chemical structures, and other scientific visuals, eliminating the need for separate molecular viewers or plotting tools .

Flexible compute management — The workbench handles job submission to local machines, SSH-accessible HPC clusters, or on-demand cloud providers (such as Modal), scaling from a single GPU to hundreds as needed .

Over 60 curated skills and connectors — Pre-configured for genomics, single-cell analysis, proteomics, structural biology, and cheminformatics, backed by more than 60 scientific databases. These connectors allow Claude to pull data from sources like Benchling and interact with platforms like BioRender .

NVIDIA BioNeMo Agent Toolkit integration — Claude Science connects natively to life-sciences foundation models in NVIDIA's BioNeMo, including Evo 2, Boltz-2, and OpenFold3, enabling tasks like protein structure prediction and molecular dynamics analysis from within the workbench .

Reviewer agent — An automated agent inspects every output for incorrect citations, untraceable numbers, and mismatches between figures and their underlying code, flagging and correcting errors in real time. This acts as a constant quality-control layer during the research process .

Basecamp Research EDEN model integration — Researchers using Claude Science can now design potent antibiotics and rapidly prioritize vaccine targets through Basecamp Research's EDEN models, generating therapeutic candidates via a conversational interface in minutes .

Availability and Pricing

Claude Science is available in beta for Claude Pro, Max, Team, and Enterprise users on macOS and Linux (installed locally or accessed via SSH/HPC login nodes) . Team and Enterprise organizations require admin enablement .

Pricing follows existing Claude plan tiers :

• Pro: $17–20/month
• Max: from $100/month
• Team: $20–25/seat/month
• Enterprise: custom pricing

Usage counts toward the same product limits as Claude Code and Claude Cowork within each plan . Anthropic also offers a discounted Team plan for active academic and nonprofit research labs .

AI for Science Grant Program

To encourage adoption in the research community, Anthropic announced a grant program offering up to $30,000 in credits for up to 50 projects, with applications open through July 15, 2026 . The program targets academic and nonprofit research teams looking to apply Claude Science to their work.

Real-World Early Use Cases

During its beta testing phase, Claude Science was used across several domains :

• Manifold Bio — Used Claude Science to nominate targets for tissue-targeting medicines by analyzing how millions of candidate binders corresponding to hundreds of targets distribute through a living body .
• Single-cell RNA sequencing, CRISPR screen design, protein structure prediction, and cheminformatics — These domains were areas where early beta testers applied Claude Science, though specific institutional details beyond Manifold Bio are not fully captured in the available source excerpts .

What Claude Science Is Not

A common point of confusion: Claude Science is not a new Claude model. It does not replace Claude Opus or Sonnet. Rather, it is a workbench application that orchestrates the existing Claude models alongside specialized scientific tools, databases, and compute resources . Anthropic has been unusually explicit about this distinction .

How It Compares to Its Predecessor

Claude Science builds on the foundation laid by "Claude for Life Sciences," announced in October 2025, which introduced connectors to Benchling and BioRender . Claude Science extends those capabilities significantly by adding local code execution, the reviewer agent, native scientific visualization, the BioNeMo integration, and the ability to manage compute across HPC and cloud resources.

Bottom Line

Claude Science represents Anthropic's attempt to do for laboratory research what Claude Code has done for software development: provide a single, auditable, AI-powered workbench that reduces context-switching and increases reproducibility. For researchers already in the Claude ecosystem, it is available immediately in beta on macOS and Linux. The AI for Science grant program lowers the barrier for academic labs to try it at scale.