Anthropic is making an aggressive bid to become the default operating system for AI-driven science, rolling out Claude Science as labs weigh productivity gains against concerns over model limits and pharma influence.
Timeline: From life-science plugins to flagship lab product
In October 2025, Anthropic quietly tested the waters with “Claude for Life Sciences,” a set of plugins that let its chatbot tap into scientific software and databases.1 On June 30, 2026, the company moved far beyond plugins, announcing Claude Science, described as an “AI workbench for scientists” that pulls fragmented tools and datasets into one research environment and can generate visuals like 3D protein structures.2
Anthropic’s own launch post bills Claude Science as an app that “integrates the tools and packages that researchers most commonly use, produces auditable artifacts, and provides flexible access to computing resources,” aimed at all stages of research from literature review to publication-ready figures and manuscripts.3 The beta is being released to Claude Pro, Max, Team, and Enterprise users, with the company promising to refine the platform based on scientist feedback.3
Human perspectives: Workflow, pharma, and reproducibility
Tech and science outlets frame the product as a workflow play rather than a new model. TechCrunch notes that Claude Science is “not a new AI model and not a more capable model for biology” but instead “bets on workflow, not a new model, to win over scientists,” giving them a single environment instead of “bouncing between databases, pipelines, and tools.”4 The Next Web similarly calls it “an app that pulls a researcher’s scattered tools into one place and lets AI agents run large parts of the work,” describing it as Anthropic’s “biggest push yet into the lab.”5
The Financial Times highlights the commercial stakes, framing the launch as a “push for pharma revenue,” with use cases in rendering 3D protein structures and drug discovery.6 MIT Technology Review goes further, calling Claude Science “Anthropic’s newest flagship product” and noting the company will use it for its own research into drugs for rare, neglected diseases—signalling both scientific ambition and potential new pharmaceutical revenue streams.1
Across coverage, reproducibility is a central selling point. Claude Science attaches an “auditable history” to each output so scientists can “validate and reproduce the results,” including the exact code and environment that produced every figure.3 The Next Web reports that “every figure arrives with the exact code and environment that produced it,” along with a plain-language note and full message history so researchers can later trace or edit results.5
AI perspective: Agents, skills, and limits
Anthropic’s own materials portray Claude Science as a coordinating agent with access to “over 60 curated skills and connectors pre-configured for genomics, single-cell, proteomics, structural biology, cheminformatics, and more,” capable of spinning up specialist agents and backed by a reviewer agent that “checks citations and calculations, flagging and correcting errors.”3
Yet reporters stress that this power sits atop the same underlying Claude models already on sale, including Opus 4.8, with “no special access and no gating.”4 The Verge underscores Anthropic’s insistence that “Claude Science is not a new AI model” and notes that the beta starts with biology but is planned to expand to other domains.2
Competitive context: Benchmarks and AI-for-science rivalry
The launch lands amid intensifying competition in AI for science. MIT Technology Review contrasts Anthropic’s move with Google DeepMind’s earlier dominance through AlphaFold and other scientific models, arguing that the “fast-advancing frontier of AI progress seems to have left DeepMind in the dust” in lucrative areas like coding.1
Meanwhile, rivals are racing to quantify AI’s ability to perform genuine research work. OpenAI president Greg Brockman recently promoted GeneBench-Pro as a way of “testing whether models can handle the kind of judgment-heavy analysis that real-world computational biology requires,” noting that its problems “would take a human expert around 20-40 hours to complete” and calling GPT-5.6 Sol “a big step forward.”
7 That benchmark underscores the same question Claude Science now faces in labs: can AI agents move beyond code execution and tool orchestration to reliably support high-stakes scientific judgment?