1 related articles
Codex's default model changed three times in 15 days, including GPT-5.5 retirement and a silent TUI default inheritance. If agent evals don't record model, reasoning effort, and account type, comparisons lose meaning. Includes a script to pin models per task.
