Anthropic announced on 28 October 2026 that its AI for Science programme is moving on from funding individual researchers to handing out free Claude credits to twenty foundations, institutes and research projects spread across several countries. The list includes public agencies, European genomics laboratories, the centre that runs the science operations of NASA's space telescopes and a community project in formalised mathematics. The company itself published the full list on its website.
Who is in, and in which group
Anthropic sorts the partners into three blocks.
The first covers those who hand money out to science: foundations and scientific societies that will receive credits to distribute among the teams they select. They include the Alfred P. Sloan Foundation, set up in 1934 to back research and training in scientific, technical and economic disciplines; the Simons Foundation, which funds mathematics and basic science and runs the Flatiron Institute; Schmidt Sciences, launched in 2024 by the Schmidts and focused on areas such as artificial intelligence, astrophysics, the biosciences, climate and space; the Open Source for Science Fund, a time-limited fund from Renaissance Philanthropy devoted to keeping alive the free software that research leans on; CAS, the division of the American Chemical Society that has spent more than a century indexing published chemistry and maintains the standard registry of chemical substances; and the United States National Science Foundation, the federal agency founded in 1950 which, according to Anthropic's own text, handles some nine billion dollars and covers roughly a quarter of federal support for basic research at the country's universities.
The second block consists of institutions that will use the credits in-house. Here we find the Space Telescope Science Institute — operated by AURA, responsible for the science operations of Hubble, the James Webb and the Roman, and keeper of one of the largest astronomical archives in the world — the Wellcome Sanger Institute, London's Francis Crick Institute with its hundred-plus research groups, the Department of Life Sciences at Imperial College London, the European intergovernmental laboratory EMBL — six sites and, through its bioinformatics institute, open databases in daily use such as Ensembl, Europe PMC or the European version of the Protein Data Bank — and the Gladstone Institutes in San Francisco.
The third block is made up of specific projects. FlyWire, at Princeton, which built the first complete map of the connections in an adult fly's brain. LEAP, an NSF centre at Columbia that blends machine learning and physics to sharpen how climate models represent clouds, oceans and soil. AstroPath, at Johns Hopkins, which applies calibration techniques from astronomical surveys to imaging tumours cell by cell. gnomAD, at the Broad Institute, which aggregates DNA from more than 800,000 people into an open reference of human genetic variation used in diagnosing rare diseases. Tau Ceti, a community project building a self-formalising mathematical library in Lean, the language that lets a machine verify every proof. And the Australian Centre for Population Genomics, a joint initiative of the Garvan Institute and the Murdoch Children's Research Institute, aimed at getting the country's least represented communities into genomic resources.
Where it comes from
The programme started in May 2025 with a simple idea: give free access to Claude to scientific projects with impact. Since then, according to Anthropic, it has supported hundreds of pieces of work in areas ranging from biology and medicine to physics, chemistry, climate and mathematics, and has opened a discounted Team plan for scientific groups.
The intermediate step was two collaborations that the announcement cites as examples. With HHMI's Janelia research campus, Anthropic has worked on the Model Hardware Standard, a shared specification so that AI agents can handle laboratory instruments safely; it is at the research preview stage. With the Allen Institute, which publishes large open datasets on the brain, cells and immunology, the company says teams of agents are being assembled that can cut analyses that used to take months down to hours.
What it means
For anyone who builds or maintains software, the interesting part is not the handing out of credits but the promise of public goods: Anthropic claims that a good share of these collaborations will produce open datasets, free software, standards and shared libraries. If that holds, what is left behind will take the form of reusable code and specifications, not a press release.
The Model Hardware Standard is the clearest candidate to have legs outside the laboratory. It is an attempt to put rules around something that is already happening — agents that do not merely write text but operate machines — and it fits the line Anthropic has been drawing, such as the rewrite of its Usage Policy with specific rules for agents connected to physical systems. Having a public specification with a name of its own gives you something to point at when it comes time to audit who moved what.
Tau Ceti and gnomAD are the other two to keep an eye on. A machine-verified mathematics library is infrastructure for anyone working on formal verification; an open genomic reference is infrastructure for half the biomedical sector. In both cases the value to third parties lies in the artefact, not in the credits.
What the announcement does not say
There are no figures. Anthropic does not spell out how many credits each partner receives, for how long, or on what criteria they are shared out within each foundation. Nor is it explained what happens once they run out: whether there is renewal, preferential pricing or nothing at all.
The claim about the Allen Institute — months of manual analysis turned into hours — comes from the company, not from a published or independently checked measurement. The same goes for the description of what the scientists say they are doing with Claude.
The data regime is not clarified either: nothing is said about whether research material passing through Claude may be used for training, nor about where it is processed, a detail that is hardly minor when we are talking about human genomics or clinical records. And the headline mentions twenty partners, yet the three blocks name eighteen; the remaining two appear to be Janelia and the Allen Institute, cited separately as collaborations already under way.
The announcement closes with a statement of intent: many of these collaborations are only beginning and the company says it wants to add more partners in future. The results, if they come, will show up in the repositories.