OpenAI Is Giving 100,000 Researchers Free ChatGPT Access, Anthropic is Taking a Different Approach
- David Borish
- Jul 31
- 5 min read
OpenAI announced ChatGPT for Academic Researchers this week, a program that will give 100,000 researchers at selected universities free access to its frontier models through 2027. The rollout starts small: 10,000 researchers this summer, with access already live at the Institute for Advanced Study and École normale supérieure. Each approved researcher can invite up to four collaborators from their own institution, and participants get a year of access to GPT-5.6 Sol Pro across ChatGPT, ChatGPT Work, and Codex, along with expanded deep research, higher usage limits, and larger context windows.
The timing is notable. Anthropic launched Claude Science on June 30, its own AI workbench aimed at researchers, along with a much smaller grant program: up to 50 projects, each receiving up to $30,000 in credits, with Modal adding up to $2,000 in compute for select recipients. Applications for that program closed July 15, with awards announced by July 31. Set against OpenAI's 100,000-seat commitment, the scale difference is stark, but the two programs are not really built to do the same job.
What OpenAI Is Offering
ChatGPT for Academic Researchers sits inside a broader OpenAI commitment of more than $250 million through 2027 for external scientific research, a figure that also covers the $50 million NextGenAI consortium and OpenAI's work with the Department of Energy's Genesis Mission to bring frontier models to national laboratories. Eligible researchers need to be faculty or postdocs at recognized, degree-granting institutions with substantial research activity, and each accepted applicant's workspace comes with business-grade privacy protections and data excluded from model training by default.
The tool access itself is broad rather than specialized. Researchers get more than 75 life science skills covering genetics, genomics, sequencing, single-cell analysis, protein modeling, and drug discovery, plus connectors to scientific literature databases, public genomic and clinical repositories, satellite imagery, computational notebooks, and reference managers like Zotero. Codex handles code writing, debugging, and reproducible pipeline work, while ChatGPT Work is positioned for longer projects such as grant applications, literature reviews, and manuscript drafts.
OpenAI backs the program with its own usage data, and this is where some caution is warranted. The company reports roughly 1.3 million people using ChatGPT weekly for advanced science and mathematics, generating about 8.4 million messages, and points to a rise in ChatGPT acknowledgments in arXiv mathematics papers, from 14 in February to 100 in the first three weeks of July. On its FrontierMath Tier 4 benchmark, OpenAI reports GPT-5.6 Sol scoring 83 percent, up from 72.5 percent for GPT-5.5, and 31.5 percent on GeneBench Pro, a biological data analysis benchmark. All of these figures come from OpenAI's own measurement and benchmark suite rather than independent replication, so they describe what the company is seeing on its own platform rather than a third-party audit of research quality.
Anthropic's Answer: A Workbench, Not a Seat Count
Claude Science takes a narrower approach. Rather than distributing broad model access to a large researcher population, Anthropic built a dedicated application that folds the scattered tools of a typical research workflow, PubMed, Jupyter, R, a cluster terminal, and dozens of specialized databases, into one environment. According to Anthropic's own announcement, the app runs wherever researchers already work, locally on macOS or Linux, over SSH to a lab's own machine, or through an HPC login node, so sensitive datasets never have to leave the systems they are already stored on.
The workbench includes more than 60 curated skills and connectors covering genomics, single-cell analysis, proteomics, structural biology, and cheminformatics, built on an integration with NVIDIA's BioNeMo Agent Toolkit that connects to life-science models including Evo 2, Boltz-2, and OpenFold3. A separate reviewer agent checks citations and calculations as the work proceeds, flagging numbers that cannot be traced back to their source code or figures that do not match their underlying data. Every output carries what Anthropic calls an auditable history, the code, environment, and message history that produced it, intended to make results easier to validate and reproduce later.
Anthropic points to three early adopters rather than aggregate usage figures. Manifold Bio, a company designing tissue-targeting medicines, used Claude Science to rank candidate drug targets against criteria drawn from its own internal data. Jérôme Lecoq, a neuroscientist at the Allen Institute, built a multi-agent pipeline of roughly 20 custom skills to write long-form literature reviews, cutting a process that used to take his team up to two years down to a fraction of that, with about ten reviews completed so far, several running past 100 pages. Stephen Francis, an epidemiologist at the UCSF Brain Tumor Center, reported that germline genetic workups on glioma susceptibility that once took his lab weeks now take roughly a tenth of the time, a result his group independently validated. These are self-reported case studies from Anthropic's own announcement rather than peer-reviewed outcomes, and they describe individual labs rather than a broad sample.
Claude Science is currently in beta for Pro, Max, Team, and Enterprise plans, with Team and Enterprise access requiring an administrator to turn it on. Anthropic is also offering a discounted Team plan specifically for academic and nonprofit research labs, verified through a lab's principal investigator.
Different Bets on How Adoption Happens
The contrast between the two programs comes down to what each company thinks the binding constraint on AI-assisted research actually is. OpenAI's bet is that free, high-limit access to a general frontier model, distributed to as many researchers as possible, will get the tool embedded into how a generation of scientists works before habits calcify around a competitor. The 100,000-seat target, the four-collaborator invite structure, and the direct outreach to institutions like IAS and ENS all point toward building a large, durable user base inside academia.
Anthropic's bet, at least with Claude Science, is narrower. Instead of maximizing the number of researchers with access, it is trying to reduce the number of tools a working scientist has to juggle inside a single session, and to make every output traceable back to the code and data that produced it. The AI for Science grant program funds far fewer projects, 50 compared to OpenAI's tens of thousands of seats, but each grant comes bundled with compute (Modal credits) rather than model access alone, and the early focus is deliberately narrow: biology and biomedical research, though the program says other domains are welcome to apply.
Neither approach has been tested at scale long enough to say which produces better research outcomes. OpenAI's numbers describe engagement and benchmark performance. Anthropic's numbers describe time saved on specific workflows at specific labs. Both are the kind of evidence a vendor would highlight, and neither substitutes for independent evaluation of whether either tool changes what gets published, replicated, or funded.
What This Means for Researchers Choosing a Tool
For a researcher deciding where to spend limited time learning a new system, the practical differences are concrete. OpenAI's program is a better fit for someone who wants broad frontier model access across writing, coding, and literature review without restructuring their existing toolchain, and who is at one of the initial partner institutions or expects to be added as the rollout expands through 2027. Claude Science is built for labs whose primary friction is switching between disconnected tools, particularly in genomics, structural biology, or cheminformatics, and who want compute management and reproducibility baked into the workflow rather than assembled by hand.
The funding structures also point in different directions. OpenAI's $250 million commitment through 2027 is oriented toward broad institutional access and government lab partnerships through the Genesis Mission. Anthropic's $30,000-per-project credits, capped at 50 awards for this cycle, function more like a targeted grant than an infrastructure rollout, aimed at proving out specific use cases in biology before any wider expansion.
Both companies frame these programs as accelerating scientific discovery. The arXiv acknowledgment counts, the FrontierMath scores, and the individual lab case studies are the evidence each company has chosen to publish so far. Whether either tool shifts the pace or quality of published research, rather than just the pace of adoption, is the question the next year of use will need to answer.
