Toby Ord quantifies swarm scaling: 16x parallel sampling equals just 4.9x longer chain-of-thought
tobyordoxford · x · 2026-09-22
Philosopher Toby Ord used benchmark data published on OpenAI's GPT-5.6 launch page to quantitatively compare two scaling paths: widening parallel sampling (swarm scaling) versus lengthening the chain of thought.
- His earlier derivation put the swarm scaling curve at 57% as steep as the serial curves, i.e. an exponent of λ = 0.57.
- That implies scaling the swarm size by 16x yields the same performance gain as extending the chain of thought by only 16^0.57 ≈ 4.9x.
- He notes OpenAI's GPT-5.6 launch page provides three benchmark data points sufficient for this kind of fit.
The takeaway: per unit of compute, longer reasoning chains currently buy more performance than fanning out parallel samples.
More from Research
- minpts-check: one-run predictor tells if your HDBSCAN clusters survive min_samples changes — leland_mcinnes · 2026-11-12
- Architecture study sparks monitoring concerns over a multi-hop planning spike — ChowdhuryNeil · 2026-11-12
- SGLang Megakernel Pushes Muse Spark to 1,109 tok/s, 2.1x Over Baseline — hsu_byron · 2026-11-12
- ArcGrav: a cross-model Agent Control Language to cut token overhead — ArchitecturalGravity · 2026-11-12
- Open-source dual-arm robot solves tangram puzzles with 90 demos, ~60% success — DanGrover · 2026-11-12
- FutureHouse's deslopper trains Kosmos to write indistinguishably from human experts — proteinrosh · 2026-11-12