Each result took just 3 hours of ChatGPT Pro thinking compute, researcher says
Dr_Atoosa · x · 2026-10-07
Researcher ConstantinKogl highlights the key sentence from a GitHub repo: on average, each result used three hours of ChatGPT Pro thinking compute.
He reveals that an important result he'll publish on arXiv tomorrow with a collaborator was found in a session where the model ('Astra') thought for 6 hours total, with the main problem-solving step done in 2.5 hours. If results at this level are possible with one-shot solutions in three hours, he calls it "insanely impressive" — a striking data point on how far long-horizon reasoning compute can go in real research.
More from Models
- Claude Opus 5.5 Rebuilds a 2-Year-Old Tool Into a Polished Industrial Monitoring Panel in Minutes — karminski3 · 2026-10-07
- Elliot Glazer: OpenAI drop lived up to hype, but almost all month-long rumors turned out false — ctjlewis · 2026-10-07
- Self-taught physics learner finds local Qwen and Ministral too weak, asks for alternatives — Available_Pressure47 · 2026-10-07
- OpenAI reasoning models went from basic arithmetic to decades-old math breakthroughs in two years — daniel_mac8 · 2026-10-07
- Gary Marcus: OpenAI's Math Proofs Rely on Symbolic AI, Vindicating His Stance but Not AGI — GaryMarcus · 2026-10-07
- Anthropic investigating elevated errors on Claude usage dashboard, Messages API unaffected — ClaudeAI-mod-bot · 2026-10-07