Anthropic Expands AI for Science: 20 Foundations and Institutions Get Claude Credits

Anthropic has expanded its AI for Science program from helping individual researchers to partnering with whole foundations, institutes and projects. In an announcement dated October 28, 2026, the company highlights 20 organizations, from the US National Science Foundation to the Wellcome Sanger Institute, that will receive free Claude credits and pass them on to scientists or use them inside their own labs.

This is a program update, not a model launch, and it carries no benchmark. What makes it worth reading is the shape of the list: it shows where Anthropic thinks scientific AI usage will concentrate, and which open datasets and standards it hopes to help grow. This post walks through the partners, what the credits are for, how this fits earlier Anthropic science moves we have covered, and what remains unproven.

TL;DR: AI for Science expansion in one table

What is the AI for Science program?

Anthropic AI for Science illustrated as a flask with rising dots, standing for research accelerated by Claude

Anthropic launched AI for Science in May 2025 on the belief that AI could significantly speed up scientific progress. The mechanism is simple: free Claude credits for researchers working on high-impact projects. The stated goal is to widen access to AI tools so scientists can "deepen and expand their own scientific work", rather than to have the lab do the science for them.

According to the announcement, scientists use Claude to generate hypotheses, design experiments, analyze complex data quickly and communicate findings. Anthropic also says that over the last few months it launched a discounted Team plan for scientists (covered in our post on the Claude Team plan for scientists) and has supported hundreds of projects.

The October 28 update changes the unit of support. Instead of only giving credits to one researcher at a time, Anthropic now gives them to organizations that can multiply them: funders that run grant programs, institutes with large teams, and open projects that serve many labs.

Who is on the list?

Anthropic splits the 20 names into three groups.

Science funders and societies

These organizations receive credits that they will distribute to ambitious researchers aiming to use AI for scientific research.

The Simons Foundation name will be familiar if you read our earlier piece on the $15M Simons effort for mathematicians with OpenAI and Anthropic. The NSF entry sits alongside Anthropic's separate federal work, covered in Anthropic's $150M Genesis Mission commitment.

Institutions

Institutional partners will use Claude credits inside their organizations.

• Space Telescope Science Institute (STScI): runs science operations for Hubble, James Webb and Roman, and holds one of the largest astronomy data archives.
• Wellcome Sanger Institute: a leading genomics center spanning human genetics, biodiversity and drug development.
• The Francis Crick Institute: one of Europe's largest biomedical institutes, with more than 100 research groups.
• Imperial College London, Department of Life Sciences: research from single molecules to ecosystems.
• EMBL: Europe's intergovernmental life science lab with six sites; its EMBL-EBI hosts Ensembl, Protein Data Bank in Europe and Europe PMC.
• Gladstone Institutes: a San Francisco biomedical institute working on cardiology, neurology, infectious disease, cancer and autoimmune disease.

Research projects and initiatives

Anthropic also gives credits directly to initiatives it says have large public-impact potential:

• FlyWire (Princeton, neuroscience): produced the first complete wiring map of an adult fruit fly's brain and builds open connectomics tools.
• LEAP (Columbia, Earth system): an NSF center combining machine learning with physics to improve how climate models represent clouds, oceans and land.
• AstroPath (Johns Hopkins, cancer research): uses calibration methods from astronomical surveys to image tumors cell by cell with consistent measurements across slides and labs.
• gnomAD (Broad Institute, genomics): pools DNA data from over 800,000 people into an open reference of human genetic variation used for rare disease diagnosis.
• Tau Ceti (mathematics): an open, community-driven project building a shared autoformalized library of mathematics in Lean.
• Centre for Population Genomics (Garvan and Murdoch Children's): increases the inclusion of Australia's multicultural communities in genomic resources.

Why this list is more interesting than a typical partner announcement

Figures gathered around an open box, a picture of the open datasets and software Anthropic says its science partners will share

Look at what these groups have in common: most of them steward shared infrastructure. gnomAD, FlyWire, EMBL-EBI's data resources, the CAS Registry, Tau Ceti's Lean library and the Open Source for Science Fund all sit underneath other people's research. Credits given to a funder or a data resource can spread across many downstream users. That is a different bet from giving a single lab a chat subscription.

It also explains the stress on public goods. Anthropic writes that many of the projects will produce "open datasets, open-source software, open standards, and shared libraries" that benefit the wider scientific community. If that happens, the value outlasts the credits themselves, because a cleaned dataset or a verified library stays useful after the usage allowance runs out.

There is a link to formal mathematics too. Tau Ceti's autoformalization effort is a direct match for the work we covered in Physical Superintelligence's open quantum problems with Lean proofs, where a formal proof checker is what separates a verified claim from a plausible one.

Agents in the lab: Janelia and the Allen Institute

The announcement gives two examples of deeper collaboration that predate this expansion.

1. HHMI Janelia Research Campus and the Model Hardware Standard. Anthropic worked with Janelia to develop the Model Hardware Standard, "a shared specification that lets AI agents safely operate lab instruments." It is in research preview. Anthropic's own MHS post has the technical detail, and we summarized it in Anthropic Model Hardware Standard: MCP for Physical Lab Equipment.
2. The Allen Institute. Anthropic says it supports the institute, which creates large open datasets and tools for brain, cell and immune system research. The institute is building teams of AI agents to analyze that data, "turning months of manual analysis into hours." That is Anthropic's characterization in the announcement; it gives no measured benchmark, so treat the speed-up as a claim.

These two examples point at where the program is heading: not just Claude as a writing and coding assistant for scientists, but agents that connect to instruments and to large open datasets. Our overview of Claude Science, Anthropic's AI workbench for scientists covers the product side of that direction, and the Guest Investigator program for life scientists covers a more hands-on collaboration route.

What this means if you are a researcher

You probably cannot apply to Anthropic for the partner credits directly, because the new money flows through intermediaries. Practical steps:

1. Check your funder. If your grants come from Sloan, Simons, Schmidt Sciences or NSF, ask the program officer whether Claude credits are attached to your award or available as a supplement. Anthropic does not describe the mechanics, so do not assume a specific allowance.
2. Check your institution. Researchers at the Crick, Sanger, EMBL, Imperial Life Sciences, Gladstone or STScI should ask their IT or research computing team whether an institutional Claude arrangement exists.
3. Use the data resources. If your field touches gnomAD, FlyWire, EMBL-EBI resources or the CAS Registry, watch those projects for AI-ready tools and shared libraries.
4. Keep your own verification. Credits buy access, not correctness. Our coverage of the Into the Unknown benchmark of AI agent papers is a reminder that agent-generated research needs integrity checks before it is cited.

For readers who are not scientists, the takeaway is about direction: AI labs are competing to be the default tool inside research institutions, and the contest is being fought with credits, standards and partnerships as much as with models.

What is not known

Anthropic's post leaves several things open, and it is worth being plain about them.

• No dollar value. The announcement does not state the size of any credit grant or the total across the 20 partners.
• No usage terms. It does not say how partners may use the credits, how long they last, or what data-handling terms apply to unpublished research.
• No outcomes. Anthropic says many collaborations are "still in their early stages" and that it will share impact later. There are no discovery claims to evaluate yet.
• Selection criteria. The post says these are organizations "we've partnered with" and highlights them; it does not say how many others exist or how a funder or institute becomes a partner.

Those gaps matter because credits create a dependency. A lab that builds a workflow around one vendor's model should ask what happens when the allowance ends, and whether its pipelines can move to other models. That question applies equally to competing programs, such as OpenAI's Rosalind academic plan and Google DeepMind's offer to 200 European researchers, which also trade free access for adoption.

How it compares with other lab science programs

The pattern across all of them is the same: the labs are giving access away to become part of research workflows, and each wants to point to shared infrastructure or published results as proof it was worthwhile. The distinguishing detail of Anthropic's update is the emphasis on funders and open data stewards rather than only on named labs.

What to watch next

• Impact reports. Anthropic promises to share results; the first concrete outputs, such as an open dataset or library release, will show whether the public-goods framing holds.
• MHS adoption. If more instrument makers or institutes adopt the Model Hardware Standard, it becomes a real interface layer rather than a research preview.
• Funder mechanics. Details from Sloan, Simons, Schmidt Sciences and NSF on how researchers actually receive credits.
• Competing offers. Whether other labs match with funder-level partnerships.

We will add an update section here if Anthropic or a partner publishes terms or results.

Related reading

• Claude Team Plan for Scientists: 10,000 Free & Discounted Seats
• Anthropic Model Hardware Standard: MCP for Physical Lab Equipment
• Claude Science: Anthropic's AI Workbench for Scientists
• Anthropic Commits $150M to the Genesis Mission
• Simons Foundation, OpenAI and Anthropic: $15M for Mathematicians
• Anthropic Guest Investigator Program
• Official source: Anthropic, AI for Science expands to foundations and institutions

Details are accurate as of October 28, 2026, based on Anthropic's announcement of the same date. Program terms may change.