I often reach the same point in Codex Desktop: most of a task is done, only a few small edits remain, and I would rather finish with a faster, less expensive model.
DeepSeek V4 Flash can now connect directly to Codex through the Responses API. But when I previously changed providers, I ran into a different problem: the conversation still existed, but the original thread no longer appeared in the same place. I eventually used opencodex to manage model routing, which let me switch models and keep working in the current task. This post covers that setup and a few startup messages that are easy to misread.