ChatGPT for Academic Researchers: Free for 100,000

How the free ChatGPT for researchers program rolls out

OpenAI is releasing the program in stages, starting with 10,000 researchers in summer 2026 and growing to 100,000 through 2027. Access is already live at institutions such as the Institute for Advanced Study and École normale supérieure.

100,000 researchers through 2027, starting with 10,000

The first group is 10,000 researchers in summer 2026, and OpenAI plans to reach 100,000 through 2027. The announcement is dated 29 July 2026. Applications opened on that same day.

Early access at IAS and École normale supérieure

The Institute for Advanced Study (IAS) and École normale supérieure (ENS) already have access. OpenAI gives them as examples ("institutions such as"), so the announcement does not say these are the only two.

What free ChatGPT access includes: models and deep research

Free access covers more than accounts: participants get OpenAI's frontier models in three products, plus larger limits, research skills and training. Three terms need a short explanation first.

• Deep research is a feature where the AI investigates many sources on its own and returns one consolidated report.
• Context window is the amount of text a model can read at one time. A larger window lets it handle longer papers and datasets in a single request.
• Higher usage limits means the cap on how much participants can use the models is raised.

GPT-5.6 Terra, Luna and Sol across ChatGPT and Codex

The GPT-5.6 family is included from launch in all three products: ChatGPT, ChatGPT Work and Codex. Participants also get expanded deep research, higher usage limits and larger context windows. OpenAI says these support scientific reasoning, agentic execution (the AI carrying out multi-step tasks itself) and research workflows across disciplines.

The three GPT-5.6 models have separate jobs: Terra balances capability and efficiency for everyday research, Luna gives faster responses on lighter tasks, and Sol takes on the hardest scientific and mathematical problems. Pricing and positioning of the three models are covered in GPT-5.6 Sol, Terra and Luna: Specs and Pricing.

Benchmark results for research-level math and biology

OpenAI cites two test results: one for research-level mathematical reasoning (FrontierMath Tier 4) and one for complex biological data analysis (GeneBench Pro). In math, GPT-5.6 Sol scores higher than GPT-5.5. The announcement gives no earlier-model score for GeneBench Pro, so that figure has no comparison.

FrontierMath Tier 4 (score, %)

Higher is better. GPT-5.6 Sol scores 83 against 72.5 for GPT-5.5.

83GPT-5.6 Sol

72.5GPT-5.5

GeneBench Pro (tasks solved, %)

Share of tasks solved, higher is better. OpenAI gives no GPT-5.5 score for this test.

31.5GPT-5.6 Sol Pro

0255075

Life science skills, connectors and training for research

The package includes more than 75 life science skills. They span genetics, genomics, sequencing, single-cell analysis, protein modeling and drug discovery. Connectors link the models to reference managers, data platforms, computational notebooks, satellite imagery, and public genomic and clinical databases, as well as scientific literature.

Codex can help with writing and debugging code, analyzing datasets and building workflows that others can reproduce. ChatGPT Work can support longer projects, from literature reviews and manuscript drafts to grant applications, funding searches and materials for communicating results. The program will also offer training for different levels of experience, from getting started to advanced research applications.

Papers and grant documents often arrive as PDFs. Converting a PDF to Markdown first can make the text easier to paste into a model.

Who qualifies and how to apply, including ChatGPT Edu

Eligibility is decided at the institution level first and the individual second. A researcher whose institution is not among the selected ones cannot qualify on personal merit alone.

Degree-granting institutions with high research activity

An eligible institution has to be a recognized college or university that grants degrees and has a high level of research activity. The initial program is limited to qualifying researchers at selected institutions of that kind.

The program targets researchers in the sciences, mathematics and engineering, with uses that run from preparing grant applications to testing hypotheses. Beyond the two examples above, the announcement text does not name the selected institutions.

Application steps and the four-collaborator limit

An applicant has to verify their institutional affiliation and describe their active research and intended scientific use. After approval, the researcher may invite up to four collaborators from the same institution. Each collaborator must also verify their affiliation, and each one counts toward the program's total number of accounts.

The announcement does not say whether the approved researcher's own account is part of that count, so it is not added up here. Applications opened on the day of the announcement, 29 July 2026.

How institutions with ChatGPT Edu are handled

At an institution that already has ChatGPT Edu, free access under this program goes through the institution's workspace. A researcher at such a school should therefore ask the institution how access is handled there.

Data privacy and fine print for academic research use

OpenAI gives a short answer on data handling, and the program has a few other stated conditions. Researchers working with unpublished results may want to check the data terms first.

Training on your data is off by default

Workspaces carry business-grade privacy and security protections, and data in them is not used to train OpenAI's models by default. That is OpenAI's own statement in the announcement. It does not describe settings for any other product or plan.

The program also has three stated conditions. It is an initial program for selected institutions only, and collaborators count toward the program's total number of accounts. The 100,000 figure is a plan through 2027 that starts with 10,000 researchers in summer 2026.

Why OpenAI backs ChatGPT for scientists and math

The announcement cites OpenAI's own usage data from research work and a larger funding commitment. All figures below are OpenAI's.

1.3 million weekly users of ChatGPT for advanced science

About 1.3 million people use ChatGPT for advanced science and mathematics every week, and they send roughly 8.4 million messages. OpenAI says the change is clearest in mathematics: in the six months before the announcement, AI has moved from occasional use on single problems to a more regular part of mathematical research. A growing number of papers now acknowledge ChatGPT's contribution.

Heavy users delegate four-hour tasks almost twice as often

Researchers in the top 20% of AI usage in their field are almost twice as likely as their peers to hand AI tasks estimated at four hours or more. For these heavy users such tasks are nearly 7% of requests, against 3.5% for other researchers in the same field.

The $250 million commitment, NextGenAI and Genesis Mission

This program is part of a larger OpenAI commitment: more than $250 million through 2027 to support outside scientific research. NextGenAI, a $50 million initiative for research institutions, is included. So is work with the Department of Energy's Genesis Mission, so that researchers at national laboratories and universities can use frontier AI.

OpenAI describes its strategy this way: it will not choose which scientific problems deserve attention or try to solve them all itself. It will put capable tools in the hands of the research community and let researchers pursue the questions they know best.

Checklist before applying for free ChatGPT for research

Six points from the announcement are worth checking before applying. Each one restates a point covered above.

• Is my institution a recognized, degree-granting college or university with a high level of research activity, and one of the selected institutions?
• Can I verify my institutional affiliation?
• Can I describe my active research and my intended scientific use?
• Do I plan to invite collaborators? Up to four from my institution may join, each must verify their affiliation, and they count toward the program's total number of accounts.
• Does my institution have ChatGPT Edu? If so, access is coordinated through the institution's workspace.
• Do the data terms fit my work? Workspaces carry business-grade protections, and data is not used to train the models by default.

If you can answer yes to the first three, the next step is the application OpenAI opened on 29 July 2026. OpenAI describes ChatGPT for academic researchers as an initial program open to qualifying researchers at selected institutions.