OpenAI wants its AI in every lab. In a new study of more than 1,800 academic scientists, those surveyed said work done with ChatGPT would have taken 1.9 times as long without it. Now the company is giving free access to more than 200,000 researchers and has opened a science blog to show what its models are discovering, from math proofs to particle physics.

OpenAI Science: Key Takeaways

  • OpenAI studied more than 580,000 research messages from 1,819 scientists in its ChatGPT for Academic Researchers pilot

  • Surveyed scientists said the same work at the same quality would have taken 1.9 times as long without AI, and 98% expect AI to speed up discovery in their field

  • ChatGPT for Academic Researchers is expanding, with free access for more than 200,000 researchers

  • A new OpenAI science blog will publish results, methods and scientists’ own accounts, starting with a particle physics proof built with GPT-6 Astra

OpenAI launched the science blog on Nov. 16, saying it wants “to give scientists and academic researchers a place to share what they’re discovering with AI and how they’re doing the work.” It calls speeding up scientific discovery “one of the most impactful ways” to pursue its mission.

What The Study Found

OpenAI’s economic research team looked at how scientists used ChatGPT over a summer pilot, sorting their messages by task with automated tools rather than reading them, and surveyed 251 of the participants. Scientists used it for everything from reviewing past studies to analyzing data and writing code.

The AI is not just doing chores. About 68% of the scientists used ChatGPT to develop research ideas. In those conversations, scientists brought their own question 35% of the time, refined an idea together with ChatGPT 36% of the time, and asked ChatGPT to suggest a research question 23% of the time.

Study FindingNumberScientists studied1,819, with 580,000+ research messagesHow much longer work would take without AI1.9 times, by scientists’ own estimatesUsed ChatGPT to develop research ideas68%Expect AI to speed up discovery98% of those who answeredResearchers getting free accessMore than 200,000

Scientists who expected faster progress mostly said AI would get research done faster, but some said it would help them find more promising questions or chase a wider range of them. OpenAI admits the big question is still open: whether those hopes turn into faster discovery “will take more research,” and progress also depends on funding, staff and lab capacity.

From Math Proofs To Brain Cells

The blog’s first post comes from physicist Nathaniel Craig, who used GPT-6 Astra on one of physics’ deepest open problems: giving the Standard Model, the theory of the known particles and forces, a complete mathematical definition on a lattice, a grid representing spacetime. Craig said he had a strategy but could not prove it worked in every case. “Astra helped turn that strategy into a construction and took the lead in developing the detailed proof,” he wrote.

OpenAI also pointed to an internal system that produced a proof addressing the Navier-Stokes problem about how fluids move, a result on the quasi-Riemann hypothesis about prime numbers, and a biology project with PerturbAI and Scripps Research in which thousands of AI agents analyzed data from millions of mouse brain cells in 32 minutes. Researcher César de la Fuente‘s team uses ChatGPT and Codex to find possible new antibiotics, which they then make and test in the lab.

Also Read: OpenAI Sets 5 Rules To Stop AI Agents Going Off Script

Why OpenAI Is Betting On Science

This is a land grab for the world’s labs. Getting hundreds of thousands of researchers using ChatGPT for free builds habits, and every breakthrough credited to its models is a powerful advert for its paid products. It also gives OpenAI a stronger answer to critics who ask what AI is actually good for.

It raises hard questions too, and OpenAI names them. In math, AI-written proofs raise the question of how new arguments get checked and explained. Across science, researchers are asking what these tools mean for training, credit for discoveries and the skills young scientists need.

What To Watch Next

Watch whether outside mathematicians and physicists confirm the AI-assisted results, since proofs only count once experts check them. The productivity numbers are also self-reported, so independent studies will matter.

Then watch the rivals. Google DeepMind and others are pitching their own models to scientists, and the lab that wins over researchers early could shape how science is done for years.

Read Next: Google Treats Its Own AI Agents Like Rogue Staff After Catching Cheats