I have been primarily using Codex (high/xhigh) for reasoning/exploring , but at times trivial and repetitive work was getting delegated as well.

Like Calvin stated, we built a simple dispatch service that worked with any harness (claude, codex, etc) that could delegate to the cheaper model family trivially.

After reading this post, I cleanup the project and decided to open source it.

Post: https://modiqo.github.io/spewer/

Code is here:https://github.com/modiqo/spewer