Anthropic Expands AI for Science Program to 20 Research Partners

Anthropic said on October 28, 2026 that it is expanding its AI for Science program beyond individual researchers to 20 named partners, a mix of foundations, research institutions, and specific scientific projects that will now receive or redistribute Claude credits. The announcement, posted to Anthropic's newsroom, does not disclose dollar figures for the new partnerships, but it lays out in detail who the 20 partners are and what each is expected to do with Claude.

What Anthropic announced on October 28

The post on Anthropic's newsroom frames the expansion as a shift in how the AI for Science program distributes support. Instead of only reviewing applications from individual scientists, Anthropic is now working directly with institutions that can either use Claude credits internally across many research groups or redistribute credits to researchers in their own networks. Anthropic says many of the resulting collaborations are still in early stages and that it plans to add more partners worldwide over time.

Across the 20 partners, Anthropic groups the relationships into four rough categories: two institutional partners with deeper technical collaborations, six funders and scientific societies that will pass credits on to researchers, six research institutions that will use credits internally, and six specific research projects that received direct grants.

The AI for Science program, briefly

Anthropic first launched AI for Science in May 2025. The original program offered qualified researchers free API credits, with eligibility reviewed by Anthropic staff and outside subject matter experts based on a project's scientific merit and how much AI could plausibly accelerate it. Anthropic's own description at launch cited work in biology and the life sciences specifically, including understanding complex biological systems, analyzing genetic data, drug discovery, and agricultural productivity; the October 28 update broadens that framing to also include physics, chemistry, climate science, and mathematics, and says the program has by now supported hundreds of projects.

Neither the May 2025 launch post nor the October 28 expansion discloses a specific dollar value for the credits individual researchers receive. Anthropic has separately rolled out a discounted Team plan for scientists, which cuts its standard seat price from $20 to $0 per month and its premium seat from $100 to $15 per month for principal investigators at accredited universities and nonprofit research institutes; that plan is open to academic researchers generally and is separate from the project-specific grants described below. Extra usage credits beyond the discounted plan are still administered through the AI for Science grant process.

The 20 new partners, by category

Anthropic's post names all 20 partners. Here is how it groups them, organized by how each will use or distribute Claude access:

HHMI Janelia and the Model Hardware Standard

The deepest technical collaboration Anthropic names is with HHMI Janelia Research Campus, which co-developed the Model Hardware Standard (MHS) alongside Anthropic. MHS is a shared specification that lets AI agents operate physical lab devices, such as microscopes, liquid handlers, and robotic arms, through a standardized driver. According to Anthropic's separate research preview announcement from August 2026, the driver exposes simple read and write primitives, makes devices discoverable in a standard format, and lets labs describe a device's safety limits in natural language so an agent knows how to operate it within bounds. The standard is model-agnostic and designed to work with any agent harness through protocols such as the Model Context Protocol.

MHS is currently in a research preview limited to a first group of scientific labs and advanced manufacturers, who are helping build out safety evaluations before Anthropic open-sources the standard. Anthropic has not given a date for that open-source release. The company also notes a real limitation: MHS only works with hardware that already exposes a programmable interface, and Claude's physical and spatial reasoning still needs expert human oversight.

The Allen Institute's AI agent teams

The second institutional partner is the Allen Institute, which Anthropic describes as a leader in large, open datasets for brain, cell, and immune research. Per Anthropic's post, the Allen Institute is building teams of AI agents to analyze its own open datasets, with the stated goal of cutting months of manual analysis down to hours. Anthropic's announcement does not provide a specific dataset, paper, or timeline for this work beyond that general description.

Funders and societies that will redistribute credits

Six philanthropic and government funders joined as partners that will pass Claude credits on to the researchers and institutions they already support, rather than using the credits themselves: the Alfred P. Sloan Foundation (founded 1934, STEM and economics research), the Simons Foundation (mathematics and basic science, including the Flatiron Institute), Schmidt Sciences (founded in 2024 by Eric and Wendy Schmidt, prioritizing AI, advanced computing, astrophysics, biosciences, climate, and space), the Open Source for Science Fund run by Renaissance Philanthropy (sustaining open-source scientific software), the US National Science Foundation (a federal agency established in 1950 that, per Anthropic's post, accounts for about a quarter of federal basic research funding at US colleges and universities on a roughly $9 billion budget), and CAS, a division of the American Chemical Society that maintains the CAS REGISTRY chemical-substance database used across the chemistry field.

Research institutions using credits internally

A second set of six institutions will use Claude credits directly inside their own labs rather than passing them along: the Space Telescope Science Institute (science operations for Hubble, James Webb, and the Roman Space Telescope), the Wellcome Sanger Institute (large-scale genomics across human genetics, biodiversity, and drug development), the Francis Crick Institute (a London biomedical institute with more than 100 research groups), Imperial College London's Department of Life Sciences (research spanning single molecules to whole ecosystems, with a focus on health, disease, and climate resilience), EMBL and its EMBL-EBI bioinformatics arm (which hosts Ensembl, the Protein Data Bank in Europe, and Europe PMC), and the Gladstone Institutes in San Francisco (integrating AI and data science across cardiology, neurology, infectious disease, cancer, and autoimmune research).

Six research projects granted credits directly

The final six partners are specific research projects, each receiving credits directly rather than through an institution or funder: FlyWire at Princeton University, which produced the first complete wiring map of an adult fruit fly brain and continues to build open connectomics tools; LEAP at Columbia University, a National Science Foundation center combining machine learning and physics to improve climate models of clouds, oceans, and land; AstroPath at Johns Hopkins University, which adapts calibration methods from astronomical surveys to image tumors cell by cell with consistent measurements across slides and labs; gnomAD at the Broad Institute, which pools DNA data from more than 800,000 people into an open reference of human genetic variation; Tau Ceti, a community-driven project building an open, autoformalized mathematics library in the Lean proof language; and the Centre for Population Genomics, a joint initiative of the Garvan Institute of Medical Research and the Murdoch Children's Research Institute that works to include Australia's multicultural communities in genomic resources and build rare-disease diagnosis tools.

Anthropic frames many of these projects as likely to produce public goods beyond the research itself, including open datasets, open-source software, open standards, and shared libraries that other scientists can reuse. For a sense of how Anthropic is positioning Claude's broader model lineup that underpins work like this, see how Opus compares to rival models it competes against for enterprise and research workloads.

How researchers can actually benefit

Given the Google autocomplete signals around this story, including "program," "grants," "application," "job," "fellowship," and "credits," it's worth being precise about what is and isn't available. Anthropic's October 28 post does not link an open application form for the new institutional partnerships, since those are direct agreements between Anthropic and the 20 named organizations rather than something a researcher applies for individually. There is also no mention of jobs or fellowships in the announcement itself.

What is open to individual researchers right now, per Anthropic's own pages, is the original AI for Science credit program (reviewed case by case based on scientific merit and potential impact) and the discounted Team plan for scientists, which is available to principal investigators at accredited universities and nonprofit research institutes in the natural sciences, math, computer science, and engineering, at $0 per month for a standard seat and $15 per month for a premium seat, with promotional pricing lasting 12 months after verification. Researchers whose institution is one of the six funders listed above, such as Sloan, Simons, Schmidt Sciences, or the NSF, may also receive credits redistributed through that organization's own grant or fellowship programs rather than applying to Anthropic directly. If you're evaluating Claude against other large models for a research workflow, Anthropic's overview of its current model lineup and pricing is a useful starting point.

What happens next

Anthropic's post is explicit that many of these 20 partnerships are still in early stages, and that the company intends to keep adding partners "worldwide" rather than treating this as a closed list. No specific milestones, renewal dates, or dollar figures were attached to the announcement. The clearest near-term marker to watch is the Model Hardware Standard: Anthropic says it plans to open-source the standard once its current research preview, running with an initial group of labs and manufacturers, has produced enough safety evaluations and best practices, though it has not given a date for that release. Anthropic has also previously discussed how it evaluates model behavior and safety more broadly in its own disclosures about model testing, which is relevant context given how much autonomy AI agents are being given over lab equipment under MHS.

For now, the practical takeaway for most researchers is that Anthropic's AI for Science program has grown from an individual-application model into a mixed system: some credits now flow through institutions and funders rather than only through direct applications, while the original individual-researcher track and the discounted Team plan for scientists remain open in parallel.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage