Yeah I’ve been using gpt-5.3-codex-spark in Codex lately and it can be surprisingly good and it’s super fast. However it needs more explicit instructions.