GPT-5.6 claims proof closing 30-year gap in optimization theory
In a single 148-minute session, Pro produced a proof that closed a complexity gap in that had been open since 1996. The prompting method reused an approach from an earlier project in which the same model proved the Cycle Double Cover Conjecture (CDC). The resulting proof was formally verified using the Lean proof assistant.
The person posting is the preprint's own author, who holds a PhD in applied mathematics and teaches in IEOR at UC Berkeley; the result has not yet been peer-reviewed. The original title said "Ultra," but the author clarified the system actually used was "Sol Pro," the top tier in ChatGPT's , distinct from the separate "Ultra" tier in Codex.
Key points
- Pro produced, in a 148-minute session, a proof closing a complexity gap open since 1996
- The proof was formally verified using the Lean proof assistant
- The method reused a prompting approach from an earlier project where the same model proved the Cycle Double Cover Conjecture (CDC)
- The poster is the preprint's own author, a PhD in applied math and IEOR teaching professor at UC Berkeley; the result is not yet peer-reviewed
- The title's "Ultra" was a mistake — the actual system used was "Sol Pro," 's top tier