GPT-5.3 Codex is retired. Use GPT-5.4 Mini for coding workflows.
Buda has removed GPT-5.3 Codex from the model selector and added GPT-5.4 Mini as the practical replacement for fast coding work.
GPT-5.3 Codex has been retired from Buda. It is no longer available in the model selector, scheduled task settings, or agent model routing.
For teams that used it for coding-heavy execution, the replacement is GPT-5.4 Mini.
This is a small operational change, but it matters. Model lists should reflect what agents can actually call. Keeping a retired model in the UI creates failed runs, confusing billing records, and unnecessary switching work for the human reviewing the output.
What changed
gpt-5.3-codexhas been removed from active Buda model choices.gpt-5.4-miniis now available as an active chat model.- Free-plan routing is supported for GPT-5.4 Mini, using the same dedicated free-model gateway path as other eligible models.
- Current pricing and model comparison content now uses GPT-5.4 Mini where a live fast GPT option is needed.
Historical release notes and older blog posts remain historical. They describe what was true at the time they were published. This post is the current operating note.
Why GPT-5.4 Mini
GPT-5.4 Mini fits the jobs where Codex was commonly used inside Buda: quick repository-aware edits, small implementation passes, test-fix loops, and repeated agent turns where speed and cost matter.
It is not the model for every final decision. Stronger models still make sense for architecture review, complex debugging, and high-risk customer-facing output.
But for execution steps, GPT-5.4 Mini is the cleaner default.
What users need to do
Most users do not need to change anything. Existing “Auto” routing will avoid the retired model.
If you saved GPT-5.3 Codex in a task template, agent setting, or internal workflow note, update it to GPT-5.4 Mini.
The rule stays simple: keep humans in charge of judgment, and give agents a model that is actually available for the execution step in front of them.