Oxford researcher uses GPT-5.6 launch data to estimate how AI agent swarm capabilities scale

tobyordoxford · x · 2026-09-22

Toby Ord (Oxford) uses the small dataset OpenAI released with GPT-5.6 to estimate how capabilities scale when inference compute goes into increasing agent swarm size rather than chain-of-thought length. His 'Swarm Scaling' thread explores how the power of large agent swarms grows as more agents are added — an empirical look at an alternative axis of inference-time scaling.

Related event: Toby Ord quantifies swarm scaling: far less efficient than longer chain-of-thought(17 posts)→

Original post →

More from Research

Research channel →