Claude Science

Scroll sideways for more →

Claude Science is an artificial intelligence research workbench released by Anthropic on June 30, 2026. It is a customizable, agentic application that pulls a scientist's scattered tools, databases, and compute into one workspace, where a coordinating Claude agent can carry out multi-step research from literature review to publication-ready figures and manuscripts ^[1][3][4]. Anthropic frames the product as doing for scientific research what its coding tool Claude Code does for software engineering: given a concise, high-level instruction, the system autonomously executes meaningful work and returns auditable results ^[1][4]. Crucially, Claude Science is not a new model. It runs the same Claude models already available to every user, including Claude Opus 4.8, so the bet is on workflow rather than raw model capability ^[3][7].

What it is

Claude Science is described by Anthropic as a single environment in which researchers can conduct all stages of their work: analyzing the literature, writing and running analysis code, querying reference databases, generating figures, and drafting papers ^[1]. Rather than a chatbot that answers questions, it behaves as a research assistant that plans and executes. Users interact with a generalist coordinating agent that has access to more than 60 curated skills and connectors, pre-configured for fields such as genomics, single-cell analysis, proteomics, structural biology, and cheminformatics ^[1][3]. The coordinating agent can hand work to specialist sub-agents and to custom agents that a lab builds for its own recurring tasks ^[1][3][4].

The product builds directly on Claude for Life Sciences, which Anthropic launched in October 2025 to add scientific connectors, Agent Skills, and life-sciences prompt libraries to Claude ^[14]. Claude Science packages that groundwork into a dedicated workbench aimed at computational biology and drug development, and Anthropic positions it as a flagship release alongside Claude Code and its agentic work assistant Claude Cowork ^[4].

Launch and positioning: "workflow, not a new model"

Anthropic unveiled Claude Science on Tuesday, June 30, 2026, at an event in San Francisco for pharmaceutical executives, biotech founders, and researchers, with CEO Dario Amodei leading the presentation ^[4][5]. Amodei, a physicist by training, argued that humans have historically grappled with biological complexity "only with their minds," and expressed hope that AI could become "a general purpose technology that helps us to make sense of that complexity." He was candid about the uncertainty, adding: "We don't know for sure if that's going to work out. But I think we're seeing signs that we're seeing the beginnings of it" ^[5].

The single most emphasized point at launch was what Claude Science is not. Anthropic stated plainly that it is "not a new AI model and not a more capable model for biology," and that it "runs the same Claude models already available to everyone today," with no special access and no gating ^[3]. TechCrunch summarized the strategy as a bet on workflow rather than a new model, reflecting Anthropic's broader shift toward vertical, workflow-level products instead of selling raw model capability ^[3]. Eric Kauderer-Abrams, Anthropic's head of life sciences, put the ambition bluntly: "It represents how important this is to our mission that this is right up there with Claude Code and Claude Cowork as the next really significant product that we're releasing" ^[4].

The launch also came days after a notable hire that underlined Anthropic's science push. On June 19, 2026, John Jumper, who shared the 2024 Nobel Prize in Chemistry for AlphaFold, announced he was leaving Google DeepMind for Anthropic after nearly nine years ^[15]. His specific role was not disclosed, but the move signaled that Anthropic intends to compete seriously in AI for the life sciences ^[4][15].

How it works

Skills, toolkits, and database connections

At the center of Claude Science is a library of more than 60 curated skills and connectors ^[1][3]. These wire the agent into the reference resources scientists actually use, including UniProt, the Protein Data Bank (PDB), Ensembl, Reactome, ClinVar, ChEMBL, GEO, and PubMed ^[1]. The toolkits cover genomics, single-cell analysis, proteomics, structural biology, and cheminformatics, and the workbench can also call external systems such as NVIDIA's BioNeMo Agent Toolkit (which exposes models like Evo 2, Boltz-2, and OpenFold3) ^[1]. Under the hood this extends Anthropic's Agent Skills and connector ecosystem, which is built on the Model Context Protocol, the open standard Anthropic uses to connect Claude to external tools and data ^[1][14].

Auditable artifacts and reproducibility

Anthropic stresses that "scientific research is inherently visual," and Claude Science renders scientific artifacts natively: 3D protein structures, genome-browser tracks, and chemical structures, which researchers can annotate inline and iteratively refine toward publication ^[1]. Reproducibility is a design goal. When the system produces a figure, it saves the exact code and computing environment that generated it, a plain-language note describing how it was built, and the full message history, so a result can be traced back to its source months later ^[1][7]. A separate reviewer (or fact-checker) agent is used to flag bad citations and mismatched numbers before results are finalized ^[1][3].

Compute and data locality

Claude Science runs locally on macOS or Linux and can reach remote resources over SSH or HPC login nodes, submitting jobs to a lab's existing infrastructure and scaling from a single GPU to hundreds as needed ^[1]. It also integrates with Modal for on-demand compute ^[1]. Because it runs on the researcher's own systems, Anthropic says large or sensitive datasets "never have to leave the systems they're already on," and the agent asks for approval before accessing new resources ^[1][17].

Relationship to Claude Code and Claude's agentic stack

Anthropic repeatedly draws the analogy to Claude Code, its agentic coding tool that reached general availability in May 2025 and grew to roughly $2.5 billion in annualized revenue by February 2026 ^[7]. Like Claude Code, Claude Science can autonomously carry out meaningful work from concise, high-level instructions, using tools and delegating to sub-agents rather than only chatting ^[1][4]. It sits within a growing family of purpose-built Claude products that now includes Claude Code, the Claude Cowork agentic work platform, @Claude in Slack, Claude Design, and Claude Security ^[7]. The common thread is agentic behavior wrapped around the same underlying Claude models, tuned to a specific professional workflow.

Target fields and users

Claude Science is aimed first at computational biology, drug discovery, and pharmaceutical research operations, with adjacent use in genomics, proteomics, cheminformatics, structural biology, and quantitative fields ^[4][5]. Anthropic named early customers including Novo Nordisk and the Allen Institute ^[4]. Several concrete uses were shown at or around launch:

Scroll sideways for more →

An independent hands-on account gives a sense of scale and cost. Writing in Forbes, epidemiologist John Drake uploaded a research library of 6,576 papers and had Claude Science analyze 490 of them on zoonotic spillover, building a bottom-up "latent ontology" of 1,240 conceptual classes. It found that of 915 relationships the literature uses repeatedly, 864 had no counterpart in the standard reference vocabulary, making the working vocabulary about four times richer than official ontologies. The total compute cost was $26 ^[9].

The AI for Science projects program

To seed adoption, Anthropic opened an AI for Science projects program alongside the launch. The company said it would support up to 50 Claude Science projects, providing up to $30,000 in Anthropic credits each, plus up to $2,000 in Modal compute credits per project ^[1][3]. Applications are open through July 15, 2026, with award notifications by July 31, and selected projects run from September 1 to December 1, 2026 ^[1][3]. Anthropic said it is prioritizing postdoctoral and graduate projects that span domains, with an early focus on biology and biomedical research ^[3].

Beyond the grant program, the beta is available to all paid Claude subscribers on the Pro, Max, Team, and Enterprise plans, with Team and Enterprise requiring an administrator to enable it, and Anthropic offering a discounted Team plan for academic and nonprofit research labs ^[1][8].

Business and strategic context

Claude Science arrived as Anthropic prepared to go public. The company filed confidential IPO paperwork on June 1, 2026, and was reported to be on track to list later in the year at a valuation around $965 billion, having closed a large Series H round ^[7]. Anthropic's annualized revenue reached roughly $47 billion by May 2026, up from about $9 billion at the end of 2025, with enterprise customers making up the large majority of that revenue ^[7]. Multiple outlets framed Claude Science explicitly as an effort to expand revenue streams ahead of the IPO, moving Anthropic's enterprise business beyond coding and workplace AI and into a new pharmaceutical and scientific market ^[7][22]. In this reading, the auditable, workflow-level design is aimed squarely at regulated, high-value research organizations that will pay for reproducibility and control ^[7].

Reception and the OpenScience open-source alternative

Coverage of the launch was broad, including MIT Technology Review, which called Claude Science Anthropic's newest flagship product, and STAT News, which noted the deliberate courtship of the pharmaceutical industry ^[4][5]. Reviewers generally credited the reproducibility and auditing features while cautioning that the tool helps most with the early, computational stages of research and that human judgment and peer review remain essential ^[9][17].

The most visible response came from the open-source community. Within about a week, on July 5, 2026, Synthetic Sciences, a San Francisco startup in Y Combinator's Winter 2026 batch, released OpenScience, an Apache-2.0 licensed, model-agnostic workbench pitched directly as an open alternative to Claude Science ^[16][19]. Where Claude Science runs only on Claude models, OpenScience is designed to work with any frontier or open-weight model using the user's own API keys, including Claude, GPT, Gemini, GLM, Kimi, and DeepSeek, as well as local fine-tunes via Ollama, with switching possible per request ^[16][18]. Synthetic Sciences advertises more than 250 editable research skill packs, which it says is more than four times the roughly 60 that ship with Claude Science, plus around 30 scientific databases (UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, and Semantic Scholar) exposed as agent tools ^[16][18][20]. The project is independent and is not affiliated with or endorsed by Anthropic ^[16].

The comparison should be read carefully. The 250-plus skills figure and the "4x" claim are Synthetic Sciences' own marketing, counting editable skill files rather than an apples-to-apples measure of capability, and OpenScience is a young project. A balanced review by Labcritics concluded that OpenScience suits computationally fluent researchers who value vendor independence and editable tooling, while Claude Science suits bench scientists who prefer a natural-language, low-code workflow without model management; it also noted that OpenScience is explicitly not sandboxed and recommends container or VM isolation, whereas Claude Science adds approval steps before accessing new resources ^[17]. Synthetic Sciences was co-founded by Aayam Bansal and Ishaan Gangwani and had raised about $1.9 million by launch ^[19]. The OpenScience release drew notable attention on X and Hacker News, positioning it as an early challenger rather than a proven substitute ^[16][20].

How it compares with related efforts

Claude Science is one entry in a fast-moving contest to build AI tooling for research, part of the broader AI for science movement. Its closest competitors take different approaches:

Scroll sideways for more →

The sharpest contrast is with OpenAI's science work: OpenAI's GPT-Rosalind, announced April 17, 2026 and named after Rosalind Franklin, is a new model specialized for biology and drug discovery, offered as a trusted-access research preview to partners such as Amgen, Moderna, and Thermo Fisher ^[21]. Anthropic's counter-bet is that scientists do not need a new model so much as a better environment around the models that already exist ^[3][4].

Limitations and open questions

The central open question is verification. In science, a confident but wrong answer is worse than none, and large language models can hallucinate citations, numbers, and mechanisms. Claude Science addresses this with a reviewer agent that flags bad citations and mismatched numbers, but as one early user observed, that reviewer "is the same model checking its own work for now, not an independent source of truth" ^[9]. Its outputs still require expert validation and, ultimately, experimental confirmation and peer review ^[9][17].

Reproducibility is a genuine improvement (saved code, environment, and message history make results traceable), yet reproducibility in AI for science remains a hard, unsolved problem: proprietary models, shifting weights, and stochastic outputs all complicate independent replication, a challenge the wider AI for science field continues to wrestle with. Anthropic itself frames the underlying scientific bet as unproven; Amodei conceded the company does not know for certain the approach will work ^[5]. Finally, because Claude Science deliberately relies on existing Claude models, its ceiling is set by those models' scientific reasoning, and much of its value depends on the quality and coverage of the 60-plus connectors and the skills a lab is willing to build and maintain ^[3][7].

ELI5 (plain-language summary)

Imagine a very capable research assistant that lives on your laptop. You tell it, in plain English, what scientific question you are chasing. It reads the relevant papers, writes and runs the analysis code itself, looks things up in the big biology and chemistry databases scientists use, draws the charts and 3D molecule pictures, and drafts the write-up, then keeps a careful record of exactly how it did everything so you (or anyone else) can check and redo it later. That is Claude Science. The clever part is that Anthropic did not build a smarter brain for it. It uses the same Claude that anyone can already use, and just gives that Claude the right tools, connections, and memory to work like a scientist. A week after it came out, a small startup released a free, open-source version called OpenScience that does something similar but lets you plug in any AI model you like.

See also

• AI for science
• OpenAI for science
• Anthropic
• Claude
• Claude Code
• Claude Opus 4.8
• Agentic AI
• Model Context Protocol
• AlphaFold
• Dario Amodei

References

1. Claude Science, an AI workbench for scientists - Anthropic (June 30, 2026) ↩
2. Claude Science (beta) - product page, Claude by Anthropic
3. Anthropic's Claude Science bets on workflow, not a new model, to win over scientists - TechCrunch (June 30, 2026) ↩
4. Claude Science is Anthropic's newest flagship product - MIT Technology Review (June 30, 2026) ↩
5. Anthropic releases Claude Science, a product aimed at researchers, the pharma industry - STAT News (June 30, 2026) ↩
6. The moment Anthropic convinced me it's serious about science - STAT News (July 1, 2026)
7. Anthropic launches Claude Science in bid to expand revenue streams ahead of IPO - Yahoo Finance (June 30, 2026) ↩
8. Anthropic Launches Claude Science Workbench For Researchers - Dataconomy (July 1, 2026) ↩
9. Anthropic's New AI Workbench Mapped My Field For $26 - Forbes / John Drake (June 30, 2026) ↩
10. Anthropic launches Claude Science, an AI workbench for scientific research - The New Stack (June 30, 2026)
11. Anthropic Launches Claude Science for Research Workflows - Enterprise DNA (2026)
12. Anthropic Launches Claude Science AI Workbench for Scientific Research - HPCwire / AIwire (June 30, 2026)
13. Anthropic launches Claude Science, an AI workspace built specifically for researchers - The Decoder (2026)
14. Claude for Life Sciences - Anthropic (October 2025) ↩
15. Nobel laureate John Jumper is leaving DeepMind for rival Anthropic - TechCrunch (June 20, 2026) ↩
16. Synthetic Sciences Releases OpenScience: An Open-Source, Model-Agnostic AI Workbench - MarkTechPost (July 5, 2026) ↩
17. OpenScience vs. Claude Science: Two Takes on the AI Science Workbench - Labcritics (July 7, 2026) ↩
18. 100% Free Open-Source Alternative to Claude Science: DeepSeek and GLM Support - 36Kr (July 2026) ↩
19. Synthetic Sciences: Foundation Models for Scientific Research - Y Combinator (W26) ↩
20. Show HN: Open Science, open-source alternative to Claude Science - Hacker News (July 2026) ↩
21. Introducing GPT-Rosalind for life sciences research - OpenAI (April 17, 2026) ↩
22. Anthropic launches 'Claude Science' to expand enterprise business and pharma revenue ahead of IPO - The Tech Portal (June 30, 2026) ↩
23. synthetic-sciences/openscience - GitHub repository