How I am splitting planning and implementation models

One thing that’s changed in my workflow recently…

With the 5.6 release of Codex, I’ve started running Medium way more than I ever did before.

On 5.5, I used Extra High for almost everything unless the task was extremely well-defined. In my experience, 5.6 Medium feels pretty comparable to what I was getting from 5.5 Extra High, so I’ve changed how I split the work.

Right now my workflow looks something like this:

• 5.6 Extra High for planning, architecture, and code reviews. • 5.6 Medium as the primary implementation model.

That combination has been giving me a really good balance of quality, speed, and cost.

I’ve also seen quite a few people having success with Luna on Max. I spent some time experimenting with it, but for my workflow I haven’t been able to get results that match what I’m seeing from 5.6 Medium. Your mileage may vary depending on the kinds of problems you’re solving.

Of course, this will probably all be outdated by next week when another model drops.

Just sharing a quick report from the front lines. 😄