"GPT-5.6 Sol Ultra produced a proof of a mathematical conjecture that had remained unsolved for 50 years, using 64 subagents in just one hour. The crucial point is not only the proof itself, which now needs to be independently verified. It is also that the model is publicly available...." — Chubby
SUMMARYGPT-5.6 Sol Ultra allegedly produced a proof of a mathematical conjecture that had remained unsolved for 50 years, using 64 subagents in about one hour. The proof still needs independent verification, and the model is publicly available, raising the possibility of more rapid scientific discovery through coordinated reasoning agents.
GPT-5.6 Sol Ultra produced a proof of a mathematical conjecture that had remained unsolved for 50 years, using 64 subagents in just one hour.
The crucial point is not only the proof itself, which now needs to be independently verified. It is also that the model is publicly available.
We are moving from AI that solves known problems toward AI that may generate entirely new scientific knowledge.
The real dimension of scaling may not simply be larger models, but coordinated swarms of reasoning agents working in parallel like a research team.
Virtually unlimited knowledge is now available to everyone, and scientific breakthroughs will become increasingly frequent. How could anyone not be excited about the future? — Chubby So it’s no longer an issue in model capability. Now the people aren’t smart enough to extract the max out of a model.
The same thing can either summarise emails OR prove an unsolved mathematical conjecture. — orcus108 Yes, you're right. But I'm an optimist. Those who want to learn new things and get the most out of the models now have the opportunity to do so. — Chubby
Source: https://x.com/kimmonismus/status/2075673485117202794