OpenAI announces a program to put frontier models in the hands of 100,000 scientists, mathematicians, and engineers at no cost. The initial launch starts with 10,000 researchers this summer at institutions like the Institute for Advanced Study and the École normale supérieure, and will roll out through 2027.
What the program offers
The goal is clear: democratize advanced AI tools to speed up scientific discovery. What does that mean for you? Participants get free access to frontier models through ChatGPT, ChatGPT Work, and Codex, with options tailored for collaboration and institutional privacy.
- Initial access for 10,000 researchers; target of 100,000 by 2027.
- Models available from the
GPT-5.6family (Terra, Luna, Sol), includingGPT-5.6 Sol Proat launch. - Invitation for up to four collaborators per researcher, all verified by the institution.
- Workspaces with enterprise-grade privacy and security protections; by default, data is not used to train the models.
- Training, hands-on support, and opportunities to share experiences among researchers.
Types of support and tools
You'll be able to use tools that range from genomic analysis and protein modeling to literature review, grant writing, and running reproducible code with Codex. Practical, right? Think of it as adding an assistant that helps with routine heavy-lifting so you can focus on the key decisions.
- More than 75 skills in life sciences: genetics, sequencing, single-cell analysis, protein modeling, drug discovery.
- Connectors to scientific literature, public genomic databases, satellite imagery, computational notebooks, data platforms, and reference managers.
ChatGPT Workfor long projects: finding funding, preparing proposals, drafting papers, and outreach materials.
Capabilities and performance
OpenAI presents three configurations designed for different trade-offs between capability and speed: GPT-5.6 Terra (balanced), GPT-5.6 Luna (faster for lighter tasks), and GPT-5.6 Sol (tackles complex scientific and mathematical problems).
- On FrontierMath Tier 4,
GPT-5.6 Solscores 83% versus 72.5% forGPT-5.5. - On GeneBench Pro,
GPT-5.6 Sol Prosolves 31.5% of tasks in complex biological analysis.
These numbers show improvements, but they are not a guarantee of peer-reviewed results. The intention is to assist you as a researcher, not replace human validation.
Why this matters
OpenAI is budgeting over $250 million through 2027 to support external research, including the $50 million NextGenAI initiative and collaborations with the Department of Energy on the Genesis mission. The idea is simple: more researchers with powerful tools accelerate science.
Can you imagine an assistant that iterates hypotheses, drafts experiment plans, and generates reproducible code while you supervise the work? That mix of speed plus human oversight is the promise here.
Usage metrics already show adoption: about 1.3 million people use ChatGPT for advanced science each week, with roughly 8.4 million messages. In mathematics, usage has moved quickly from spot help to regular integration in research, and even papers explicitly cite ChatGPT for technical contributions.
Real research examples
- A physics team uses AI to develop open research software for fusion, useful for labs and industry alike.
- Computer science researchers used
GPT-5.5 Proto propose and refine a proof about limits in high-dimensional geometry problems, then validated the results independently.
These cases show AI can push research agendas toward more ambitious questions, as long as researchers keep control and rigor.
How to participate and requirements
The call is open to researchers at selected academic institutions. Key requirements:
- Eligible institutions: recognized universities or colleges that grant degrees and have high research activity.
- Applicants must verify their institutional affiliation and describe their active research and intended use.
- Invited collaborators (up to four) must also verify affiliation and count toward the total slot.
- For institutions with ChatGPT Edu, access is coordinated through the institutional workspace.
Limits, responsibility, and best practices
Technology moves fast, but it is not infallible. OpenAI acknowledges model limits; that is why the program includes training and feedback channels.
Data is not used to train the models by default, and the program prioritizes privacy and security controls.
Even so, it is essential to verify results, replicate analyses, document workflows, and consider ethical and regulatory frameworks in your field. AI should be a tool that amplifies research, not a black box that replaces expert judgment.
Final reflection
If you are part of an academic institution, this program can be a doorway to experimenting with frontier AI on real tasks: from drafting hypotheses to accelerating complex analyses. Could AI be the partner that saves you weeks or months on a project? Possibly. Will it replace careful peer review and human validation? No.
The practical invitation is this: use these tools rigorously, share what works and what doesn’t, and keep research focused on important, verifiable questions.