Multi-agent scaling may dominate next, calls for OpenAI to publish curves

1a3orn · x · 2026-09-22

The author argues OpenAI, seemingly far ahead, should publish concrete and detailed scaling curves, as this wouldn't compromise their techniques. He notes this implies an even greater pull toward multi-agent scaling, and that current results (5.6 Sol, one of the first multi-agent trained systems) are probably nowhere near the frontier of what's possible.

Original post →

More from AGI Musings

AGI Musings channel →