OpenAI Announces GPT-5.6 Sol Ultra Successfully Solved 50-Year-Old Math Puzzle in One Hour

OpenAI announced that its latest AI model GPT-5.6 Sol Ultra completed a complete proof of the so-called Cycle Double Cover conjecture in one hour. This conjecture was independently proposed by many mathematicians in the 1970s and has troubled the academic community for about 50 years. There is no generally accepted proof yet. The model uses 64 subagents working in parallel to ultimately generate the proof. The proof was affirmed by mathematicians. Thomas Bloom, a mathematician at the University of Manchester, praised the proof as "simple and basic, and could have been discovered in the 1980s." However, he criticized the paper for not citing the relevant literature of Bermond, Jackson and Jaeger in 1983. Bloom suspects that AI is not created out of thin air, but rather reorganizes ideas from existing research, which once again triggered the discussion of whether AI can truly "create." AI shows patience that humans lack. Bloom pointed out that AI will not give up due to initial failure and will continue to try small changes until it finds a breakthrough. Human mathematicians often give up when their intuitive methods fail, and AI’s “persistence” is the key. He believes that this kind of proof originally only required existing theories plus a lot of patience and faith, and humans could have discovered it earlier. Prompt: The project must be strictly controlled. OpenAI has designed special prompts for this proof: the model is required to assume that the proof exists, it is forbidden to search the network to confirm whether it has been solved, and it is set to run for at least 8 hours before giving up. Most of the 64 subagents are deliberately kept isolated from information, and adversarial agents check for typical errors, such as closed paths being mistaken for loops. The model ultimately completed the task in less than an hour.