This may now be resolved.
See: Codex Desktop repeatedly shows “Selected model is at capacity” during active tasks - #8 by OpenAI_Support