Swarm scaling is inefficient: matching a 100x token scale-up needs 900x–15000x more agents
tobyordoxford · x · 2026-09-22
Toby Ord (Oxford) shares estimates converting swarm scaling to thinking-duration scaling: scaling the number of agents in a swarm by 10x yields only 10^λ — about 3x to 5x — of the performance gain from giving one agent 10x more tokens. The shortfall compounds at scale: to match a 100x token scale-up for a single agent, you'd need to grow the swarm by 900x to 15,000x. The implication is that longer thinking by a single agent beats piling up more agents.
More from Research
- Training on production traces: single-trajectory RL may unlock continual learning — rhythmrg · 2026-09-22
- Most compute now goes to RL, letting models surpass human data limits — MarvinTBaumann · 2026-09-22
- TinyTorch: PyTorch's free curriculum to build an ML framework from scratch in 20 modules — PyTorch · 2026-09-22
- Why AI won't boost paper output for researchers who chase hard problems — kfountou · 2026-09-22
- ICML 2027 braces for 100k submissions as AI paper boom continues — CharlotteHase · 2026-09-22
- 1080 Ti beats RTX 6000 by 2.4x on dense-model inference despite 4x less bandwidth — EAccelerate_42 · 2026-09-22