← All news·2026-07-11·3 min read

GPT-5.6 Sol Proves a Theorem Humans Couldn't Crack for 50 Years

OpenAI's GPT-5.6 Sol Ultra built a complete proof of the Cycle Double Cover Conjecture — an open graph theory problem from the 1970s — in under an hour.

airesearchmathopenai

GPT-5.6 Sol Ultra, developed by OpenAI, solved the Cycle Double Cover Conjecture (a longstanding graph theory problem since the 1970s) within an hour. It used 64 subagents working simultaneously, without realizing the problem was unsolved.

“Short and elementary — the kind that would have appeared in the 1980s” - mathematician Thomas Bloom on GPT-5.6 Sol Ultra’s proof of the Cycle Double Cover Conjecture

However, GPT-5.6 Sol Ultra’s proof doesn’t mention the seminal 1983 paper by Bermond, Jackson & Jaeger (even though it seems to be inspired by their work). This is a common problem with AI-generated proofs.

Instead of asking “Can AI do real math?”, we need to ask “What is AI actually doing?” Is it discovering novel approaches or just exhaustively trying every tool we know of? Either way, our understanding of how to use AI in science needs to change — it’s not replacing us, but finishing what we start.

Source: the-decoder.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