Anthropic expands science program to 20 research institutions with a unified standard for leading laboratory devices
Anthropic announced the expansion of its science-focused program to include twenty institutions, societies and research groups worldwide, granting researchers working on high-impact scientific projects free credits to use Cloud models. The program, which launched in May 2025 based on the hypothesis that artificial intelligence can materially accelerate scientific progress, has at this stage moved beyond individual research teams to become a umbrella that supports large entities and specialized reference centers in biology, medicine, physics, chemistry, climate science, and mathematics, after having provided in recent months a reduced-cost plan for scientific teams and support for hundreds of academic projects.
A hardware standard moves intelligent agents from analysis screens to physical testing platforms in laboratories.The company collaborated with the Janilia Institute for Research, part of the Howard Hughes Medical Institute, to develop the Model Hardware Standard (MHS), a joint specification currently under research review that allows artificial-intelligence agents to control and operate laboratory devices and equipment safely. This pathway represents a qualitative shift from merely generating hypotheses or summarizing scientific papers to delegating the agent to perform concrete experimental tasks within a lab-testing environment and to automate measurement instrument calibration.
On the open-data track, the initiative supports the Allen Institute, which is building teams of intelligent agents to analyze massive open datasets in brain, cell and immune-system research, turning manual analysis cycles that used to take many months into limited hours of work. This effort aligns with Princeton University’s FlyWire project, which produced the first complete neural-connectivity map of an adult fruit-fly brain, developing open data and tools in connectomics that enable tracing how neural circuits generate perception and behavior. In genetics, the Genome Aggregation Database (gnomAD) at the Broad Institute, which aggregates DNA data from more than 800 thousand individuals as an open reference for human genetic diversity, is used to support diagnosis and discovery of rare diseases worldwide.
A research alliance links federal and endowment funding with advanced inference software.The expansion includes long-standing partners and institutions, led by the U.S. National Science Foundation, a federal agency created by congressional act in 1950 with a budget of nine billion dollars that alone accounts for roughly 25 percent of total federal support for basic research at American colleges and universities, together with its LEAP center at Columbia University, which integrates machine learning with physics to improve climate-model simulations of cloud, ocean and land dynamics. The initiative also includes the Alfred P. Sloan Foundation, which has been supporting research and education in science, technology and economics since 1934, and Schmidt Sciences, founded by Eric Wendi Schmidt in 2024, as well as prominent entities such as the Welkom Sanger Institute, the Francis Crick Institute, Imperial College London, the American Chemical Society, the Space Telescope Science Institute, and the European Molecular Biology Laboratory.
What actually changes for researchers, academic institutions and medical centers in the Gulf, Egypt and the Levant? The ongoing shift moves scientific artificial intelligence from isolated experiments and the purchase of separate subscriptions to a direct integration of software agents with field-testing equipment. This means that universities and regional genome centers, which are heavily investing in outfitting their laboratories, now need automation engineers who can link device protocols to inference engines alongside traditional data analysts, and adopt uniform standards for the safe operation of sensitive equipment to avoid laboratory testing errors. Providing subsidized compute credits to the scientific community also requires academic leaders in the region to balance the cost of using commercial cloud inference models with the cost of hosting open models on national computing infrastructures, especially when processing large-scale health and climate data that demand long-term operational sustainability and precise local governance of sensitive data.
This step demonstrates that competition among model developers extends beyond public chat interfaces to the capture of core scientific-research environments, with free credits becoming a gateway for embedding software and hardware standards within the halls of major universities and global reference institutes.