GPT-5.6 goes GA with an "ultra" parallel-agent setting and sharp cost claims

OpenAI's GPT-5.6 release adds an "ultra" setting for spending more computation on difficult tasks. By default, ultra coordinates four agents working in parallel, which the release article says trades higher token usage for stronger results and faster time to result on demanding tasks.

That makes ultra a different operating choice, not a promise that every task becomes cheaper. The intended benefit is finishing demanding work sooner by having several agents work at once.