← All news·2026-10-27·3 min read

OpenAI Model Disproves the Exponential Time Hypothesis

An internal OpenAI model disproved the Exponential Time Hypothesis and proved Sogge's conjecture on wave regularity, and the company declined credit for the discoveries.

aiopenairesearchmath

Recently, OpenAI claimed that their internal model refuted the Exponential Time Hypothesis (ETH), a well-known conjecture in computational complexity theory. This conjecture was assumed to be true in many database and geometric related papers. In particular, on October 16th, OpenAI published that they had refuted the conjecture. Later, on October 17th, this proof was validated with the help of Lean, a tool for automated proof checking. Lean doesn’t give a damn about your beautiful proof. Either it accepts it, or it doesn’t, and that’s final! Afterwards, the same model went on to prove another mathematical conjecture called Sogge’s conjecture about the regularity of waves in any dimension. Last but not least, there is also a third result regarding four dimensional quantum field theory. Unfortunately, this one hasn’t been made public yet since OpenAI decided to let other physicists check it beforehand.

It’s important to mention that OpenAI isn’t claiming ownership over these findings. They actually gave the Simons Foundation (a research institute focused on fundamental sciences and math) $5 million so that independent mathematicians can validate these proofs. Although we cannot use this approach when programming with AI agents (at least not today), it’s nice to see where things are heading. Remember, trusting an AI agent to produce bug free code is completely different than relying on a mathematical model to prove theorems.

Source: openai.com

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids