Vals AI: agent teams cost 1.8 to 5.1 times as much, with one significant gain

TL;DR

On 9 October Vals AI compared single agents and teams of GPT-6 Sol and Claude Opus 5.5 on 50 web apps built from a product spec.
Teams cost 1.8 to 5.1 times as much, and only Sol's medium-effort team's 7.3-point gain was statistically significant.
Most of the extra cost was cached input that each subagent re-sent, and each setup ran once.