💥 Today we say “hello world” from OpenAI for Science.
We’re releasing a paper showing 13 examples of GPT-5 accelerating scientific research across math, physics, biology, and materials science. In 4 of these examples, GPT-5 helped find proofs of previously unsolved problems.
We’re releasing a paper showing 13 examples of GPT-5 accelerating scientific research across math, physics, biology, and materials science. In 4 of these examples, GPT-5 helped find proofs of previously unsolved problems.
Our aim is to be measured, yet optimistic. We show, with specific examples, what GPT-5 can and cannot do today, and give a clear path for how researchers can use it to accelerate scientific discovery while keeping standards high. We believe GPT-5 already provides substantial value for scientific researchers today, and will become an even more powerful tool tomorrow.
It’s not just about the proofs of unsolved problems. There are everyday examples of acceleration, like when GPT-5 can save hours by completing a tough calculation. It’s an incredible brainstorm partner for new ideas due to the sheer breadth of science it understands. And its literature search abilities are incredible—matching concepts across disciplines and languages.
One of my favorite quotes in the paper comes from physicist Robert Scherrer: “I have accumulated a number of such unsolved interesting mathematical problems that have frustrated me over my 40-year research career. Many of these seem particularly well-suited to AI solutions. I have long waited for this moment to arrive.”
GPT-5 is not yet solving huge open problems like the Riemann Hypothesis or contributing to the Langlands Program. But the idea that we’d be talking about an LLM providing proofs of unsolved problems would have been absurd a year ago. So the fact that we’re solving small to medium unsolved problems today means we’ll solve bigger problems in the future. And the future comes fast in AI.
I want to say a huge thank you to our co-authors: Christian Coester, Timothy Gowers (@wtgowers), Mehtaab Sawhney, Robert Scherrer, Brian Spears (@bkspears9), Derya Unutmaz (@DeryaTR_), and Nikita Zhivotovskiy. Special thanks also to @SebastienBubeck, @ALupsasca, and @MarkSellke for heavy collaboration on the paper, and to the rest of our colleagues who contributed.
Also: @ALupsasca and I joined the OpenAI Podcast to talk more about the paper and where OpenAI for Science is going: check it out at .
You can read more on our blog here: openai.com/index/accelera…
And of course, the paper! It’s linked here, and should be available on arxiv.org soon. cdn.openai.com/pdf/4a25f921-e…
Looking forward to hearing your thoughts! 2026 is going to be wild.
Looking forward to hearing your thoughts! 2026 is going to be wild.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
