Swarm scaling needs squared inference to match chain-of-thought gains, analysis of OpenAI curves finds
tobyordoxford · x · 2026-09-22
tobyord Oxford analyzes OpenAI's multi-agent swarm scaling curves and finds that spending extra inference on swarm size yields only about half the score increase of longer chain-of-thought — you need to square the inference scale-up of swarm size to match a scale-up of CoT. Since OpenAI ran no clean ablation holding CoT fixed while scaling agent count, he infers it by joining points with equal tokens-per-agent across the curves. A useful empirical signal for whether inference compute should go to CoT length or more agents.
More from Models
- ML vet chrisalbon: long-form AI writing is just bad — use AI for research, write it yourself — chrisalbon · 2026-09-22
- Testing shows Jev hallucinates like other models, even failing the strawberry r-count — JeremyNguyenPhD · 2026-09-22
- Qwen3.6-35B runs surprisingly well on a single RTX 4070, dev reports — haydendevs · 2026-09-22
- Grok 4.7 high looks like a downgrade at 3D generation and eats tokens, user tests find — teortaxesTex · 2026-09-22
- User asks Claude for yuri manga with fan service, model asks how much — repligate · 2026-09-22
- Former OpenAI researcher: many ideas were left in GPT-4 as the field moved too fast — willdepue · 2026-09-22