Video unavailable. Open the source below.
NewJev Codex Effort Router
Changes coding-agent reasoning effort during a task.
Added to Jevfast
How it uses Jev
Inference: Jev judges whether the current Codex step calls for more or less reasoning effort, and that judgment changes the effort used for a GPT-6 call. The post does not disclose the state sent to Jev, available effort choices, decision threshold, or how the router applies a result. The reported cost reduction is creator-reported.
Why it is interesting
It uses a decision inside an active coding task to adjust reasoning effort between model calls, rather than choosing a model only before the task starts.