OpenAI is giving 100,000 researchers free access to its top model by 2027

[OpenAI](#/company/openai) launched ChatGPT for Academic Researchers on July 29, a program giving scientists, mathematicians, and engineers free access to its models -- starting with 10,000 researchers this summer and expanding to 100,000 by 2027. It is the company's largest free-access research initiative to date, backed by a stated $250 million commitment to external scientific research through 2027.

What's actually included

Invited researchers get access to GPT-5.6 Sol and the higher Sol Pro tier, tuned for difficult, long-running queries, along with higher usage limits and larger context windows. The program also includes an expanded Deep Research mode able to pull from hundreds of sites or focus narrowly on sources like scientific journals, plus Codex and a task-automation tool OpenAI calls ChatGPT Work. Each invited researcher can bring on 4 colleagues from the same institution with identical access. OpenAI says workspace data is excluded from model training by default, with business-grade privacy protections. Access is free, and researchers apply through a form on OpenAI's site. Early access is already live at institutions including the Institute for Advanced Study and Ecole normale superieure.

The feature mix is telling on its own. Deep Research and the expanded journal-focused search mode are aimed squarely at literature review and evidence synthesis, the part of research that consumes the most reading time for the least novel output. Codex and ChatGPT Work point at a different use case -- automating the more mechanical parts of a research pipeline, from data cleaning to running standard analyses -- rather than the hypothesis-generation or experiment-design work researchers themselves usually consider the creative core of the job. OpenAI is betting the time savings are worth the access it's giving away; whether that holds depends on tasks the company hasn't published usage data on yet.

Not the first time

The program extends a pattern rather than starting one. OpenAI committed $50 million in AI resources to universities in May 2024, has been supporting multiple U.S. Department of Energy research projects, and launched ChatGPT Edu for universities in 2024 with its own usage analytics and cybersecurity controls. ChatGPT for Academic Researchers is the largest of these by stated scale and the first built specifically around individual researchers rather than institutions as a whole. Each prior step moved OpenAI a little deeper into how universities and national labs actually work day to day -- IT-department licensing with ChatGPT Edu, project-specific compute with the DOE partnerships -- and this program is the first aimed at the individual scientist's own workflow rather than an institution's infrastructure.

Lined up against ChatGPT Edu, the 2024 program it most resembles, the differences are as informative as the similarities. ChatGPT Edu targeted universities as institutions -- a licensing relationship with usage analytics and cybersecurity controls built for IT departments to administer. ChatGPT for Academic Researchers instead targets individual scientists directly, ties them to OpenAI's top reasoning tier rather than an unspecified one, and attaches an explicit, dated scale target -- 100,000 by 2027 -- that ChatGPT Edu never publicly carried. That shift, from institution-level licensing to researcher-level access with a named model tier, is the clearest sign of how OpenAI is now framing its relationship with the academic sector.

The evaluation-first read

Free access to a lab's best available model is a real, immediate benefit to researchers who could not otherwise afford it. It is also, structurally, how a frontier lab seeds default-tool status in the next generation of scientists -- a pattern with no independent measurement yet of whether it actually accelerates discovery, only participation targets. The two things are not in tension: a program can be a genuine research subsidy and a customer-acquisition strategy at the same time, the way corporate-education discounts and free developer tiers have worked in software for decades. What's different about applying that model to frontier AI specifically is the stakes attached to which lab's tools end up as the default in a given scientific field -- data formats, workflows, and even what counts as a citable methodology can end up shaped by whichever assistant a discipline trained on first.

For now, the concrete facts are the program's own stated terms, not a verdict on its motives: 10,000 researchers onboarding this summer, a dated path to 100,000 by 2027, and a $250 million commitment attached to it. Whether that adds up to a genuine acceleration of science or a well-funded on-ramp to a future paid tier is a question the program's own next phase -- and whatever pricing terms eventually follow it -- will actually answer.