Anthropic-style Claude Code split: Jev routes tasks in 16ms
Sonnet 5.5 builds, Haiku 5.5 scans the repo, and Opus 5.5 handles architecture and review, while Jev routes tasks in about 16ms so Opus isn't wasted on routine work.
Bro, this Claude Code setup is actually wild I thought the best way to use Claude Code was to throw Opus at everything. Apparently not. Anthropic’s setup is much more interesting: → Sonnet 5.5 builds. → Haiku 5.5 scans the repo. → Opus 5.5 architects. → Opus reviews plans, failures, and releases. → JEV routes tasks in ~16ms. The crazy part? Each model has a completely different job. Opus reviews. Sonnet builds. Haiku swarms. JEV routes. Instead of wasting Opus on routine tasks, save it for big decisions. And that changes everything. You’re not just using multiple models. You’re giving each one a specific role. Plan on high. Delegate on medium. Keep Opus on call. The edge isn’t using the smartest model for everything. It’s knowing when to use it. Want the setup?
This post quotes another post: x.com/w1nklerr/status/2096964317107077567