# OpenAI Claims a GPT-5.6 Configuration Produced a Proof of a 50-Year-Old Graph Conjecture

The system reportedly used 64 parallel agents and under an hour. Mathematicians call the proof elementary but note that it is unreviewed and thinly cited.

- Published: 2026-07-17T05:29:37.841Z
- Canonical: https://polylog.news/ai/2026-07-17/openai-claims-a-gpt-5-6-configuration-produced-a-proof-of-a
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [The Decoder](https://the-decoder.com/openais-gpt-5-6-sol-ultra-reportedly-solves-a-50-year-old-math-problem-in-under-an-hour/), [Hacker News (proof PDF discussion)](https://news.ycombinator.com/item?id=48863490)

OpenAI published a paper attributing a proof of the Cycle Double Cover Conjecture, a graph-theory problem open since the 1970s, to a system it identifies as GPT-5.6 Sol Ultra, [according to reporting](https://the-decoder.com/openais-gpt-5-6-sol-ultra-reportedly-solves-a-50-year-old-math-problem-in-under-an-hour/) aggregated from the release. The conjecture, posed independently by George Szekeres in 1973 and Paul Seymour in 1979, holds that any bridgeless graph admits a collection of cycles covering every edge exactly twice. The [AI Post channel](https://t.me/aipost/7545) reports that the model deployed 64 agents in parallel and returned the argument in under an hour.

Early expert reaction is mixed in a telling way. Thomas Bloom, a mathematician, called the proof "very nice" and "elementary," saying it could have been found in the 1980s, while criticizing its lack of citations to foundational prior work. On [Hacker News](https://news.ycombinator.com/item?id=48863490), readers are still working through the PDF. Importantly, the proof has not been peer-reviewed or formally verified.

The honest read splits into two outcomes. If the proof survives formal checking, it is a concrete instance of a model closing a decades-old open problem through parallel search and verification, not merely competition-style problem solving. If reviewers find a gap, it becomes a cautionary example of a lab announcing a result before validation. What decides between these outcomes is independent verification, ideally machine-checked in a proof assistant.

## What this means

A verified machine-generated proof of a standing conjecture would move the frontier from AI solving problems with known answers to AI generating new mathematics, which raises the stakes on verification tooling, because the bottleneck shifts from generation to trusted checking. OpenAI gains a capability narrative, while the exposed party is anyone who treats a vendor proof announcement as settled before a proof assistant or referees confirm it.

## What to watch

- Whether a formal proof-assistant check (Lean or Coq) of the argument appears, which would convert the claim from asserted to verified.
- Whether OpenAI releases the full agent trace and prompts, letting outsiders judge how much was search rather than retrieval of a near-known argument.