ChatGPT Research Program Opens to 100,000 Scientists

Frontier Models, Handed to the People Who Ask the Hardest Questions

OpenAI announced ChatGPT for Academic Researchers on July 29, 2026, a program that puts frontier model access in the hands of working scientists at no cost to them. It starts small and deliberately — roughly 10,000 researchers this summer, at institutions including the Institute for Advanced Study and École normale supérieure — and scales toward 100,000 researchers through 2027. It sits inside a broader commitment of more than $250 million to external scientific research.

• 10,000 researchers this summer, expanding to 100,000 through 2027, at selected academic institutions
• GPT-5.6 Sol Pro at launch, alongside the wider GPT-5.6 family, with expanded deep research, higher usage limits, and larger context windows
• Each participant can invite up to four collaborators from their own institution into the workspace
• Business-grade privacy: workspaces carry enterprise security protections and data is not used to train OpenAI models by default

Why the Collaborator Invites Matter More Than the Seat Count

The headline number is 100,000, but the design detail worth noticing is that each researcher can bring four colleagues into the workspace. Research does not happen at the granularity of an individual. It happens in groups — a principal investigator, two postdocs, a graduate student — and a tool that only one of them can reach becomes a bottleneck rather than an accelerant.

Anyone who has watched a lab adopt software knows the failure mode. One person gets the license, becomes the person you ask, and the tool's usefulness is capped by that person's availability. Extending access to the working group instead of the named individual is a small structural choice that determines whether this lands as a genuine capability or a curiosity.

What Does GPT-5.6 Sol Pro Actually Change for Research?

The GPT-5.6 family splits by workload in a way that maps cleanly onto how research actually gets done. Terra balances capability against efficiency for everyday work. Luna is the faster, lighter option. Sol is aimed at the hardest scientific and mathematical problems, and Sol Pro is what this program leads with.

The accompanying features are arguably as consequential as the model tier. Expanded deep research, higher rate limits, and larger context windows are exactly the three constraints that make frontier models frustrating for scholarly work. A literature review that spans forty papers does not fit in a small context window. An agentic run that reads, cross-checks, and summarizes across sources needs headroom to finish. Removing those ceilings changes which tasks are worth attempting at all.

Is Grant Writing Really the Interesting Use Case?

OpenAI names grant preparation alongside hypothesis testing, and it is easy to read the first as filler. It is not. Grant administration consumes an enormous share of a working scientist's week, and it is almost entirely undifferentiated labor — reformatting the same research program into a different funder's template, tracking compliance language, assembling budget justifications.

Time returned there is time returned to the actual work, which is the same pattern we found in our AI coverage of newsroom AI workflows that give reporters time back. The productivity story in knowledge work has consistently been less about the model doing the hard thinking and more about it clearing the administrative underbrush around the hard thinking.

The Privacy Terms Are the Precondition

Research data is frequently sensitive — unpublished results, subject data, work under embargo or industrial partnership. A program without clear data terms would be unusable for a large fraction of the intended audience regardless of how good the model is. Business-grade protections with training excluded by default is what makes the offer answerable inside a university's compliance process.

This follows a broader run of science-directed AI programs we have tracked, including Anthropic's rare disease research grants offering $50K in Claude credits and Ai2's OlmoEarth platform for continent-scale geospatial inference. Different labs, same recognition: the highest-leverage place to put frontier capability is with people whose questions are already hard and whose budgets are already thin.

The honest caveat is that 100,000 seats across global academia is meaningful but not universal, and institutional selection determines who benefits. Still, a phased rollout that starts at 10,000 and grows is the responsible shape for a program like this — it gives OpenAI real usage data before the scale-up, and it gives the first cohort a version that has been sized to work.

Sources: OpenAI — July 29, 2026; SiliconANGLE — July 29, 2026; AIwire — July 30, 2026.