Since we released it in the API, I'm consistently hearing from folks at meetups that GPT-5.3-Codex is the one model to use to get actual work done and clear winner for any background agent / automation at scale. It's improved a lot already but we still have ways to go to make it more personal and engaging, but I'm glad we're seeing it breakthrough on raw coding ability. The secret is out on which model gets the best results for a given $ amount.