GPT-5.6 Sol Ultra solves 50-year-old math conjecture — but did it just recombine old work?
openai's gpt-5.6 sol ultra proved the 50-year-old cycle double cover conjecture in under an hour using 64 parallel subagents. mathematician thomas bloom validated the proof but criticized the ai's failure to cite prior literature, raising questions about whether models are truly creative or just persistent.
openai's new model, gpt-5.6 sol ultra, reportedly generated a complete proof for the cycle double cover conjecture—a problem unproven for 50 years—in under an hour by coordinating 64 subagents. mathematician thomas bloom of the university of manchester confirmed the proof's validity, describing it as elementary and short, but criticized openai's paper for omitting citations to prior work from 1983. bloom argues the ai likely found the solution not through creative insight but by executing a 'counterintuitive twist' that humans would have abandoned too quickly, reflecting a pattern where models succeed by applying extreme patience to existing theories. this development reinforces the debate over whether reasoning models produce genuine novelty or simply recombine known literature, raising ethical concerns about attribution in ai-generated research.