Codex removes GPT-5.2 and GPT-5.3-Codex on June 2
OpenAI said ChatGPT-linked Codex will drop GPT-5.2 and GPT-5.3-Codex on June 2, with GPT-5.5 becoming the default frontier model for free users. The API versions stay available, but the in-product model surface is being reduced for compute-fleet management.
TL;DR
- In thsottiaux's announcement, OpenAI said Codex will remove GPT-5.2 and GPT-5.3-Codex for users signed in with ChatGPT accounts on June 2.
- According to thsottiaux's post, GPT-5.5 becomes the default frontier model for free-plan users inside Codex after that cutoff.
- thsottiaux's announcement also says the retired Codex entries stay available through the API, so this is a product-surface change, not a full model shutdown.
- OpenAI framed the change as compute-fleet simplification in thsottiaux's announcement, while thsottiaux's follow-up added that GPT-5.2 now accounts for less than 1 percent of production usage.
In thsottiaux's announcement, the notable detail is not just that two Codex model options are disappearing, but that the cutoff only applies when you log in with a ChatGPT account. thsottiaux's follow-up adds a blunt adoption datapoint, GPT-5.2 is already below 1 percent of production usage. And dkundel's PSA suggests the broader goal is a slimmer model picker inside Codex.
June 2 cutoff in Codex
OpenAI's product change is narrow and specific. GPT-5.2 and GPT-5.3-Codex stop showing up in Codex on June 2 for people using ChatGPT login, and GPT-5.5 becomes the default frontier model for free users.
WesRoth's summary restated the same cutoff and default-model swap a few hours later, which lines up with OpenAI's own wording.
Compute fleet management
The official reason is compute-fleet management. In thsottiaux's announcement, OpenAI said it is sunsetting the two older Codex options "to simplify our Codex compute fleet management," which reads like backend consolidation more than a capability statement about either model.
That wording matters because the API versions are not going away. The reduction is happening in Codex's in-product model surface, not across OpenAI's entire model catalog.
Less than 1 percent usage
The strongest extra datapoint came in thsottiaux's follow-up, where he said GPT-5.2 is now seeing less than 1 percent of production usage and is already considered outdated relative to the models that followed it.
That helps explain why OpenAI can remove older Codex choices so quickly. reach_vb's reaction post framed GPT-5.3-Codex as a short-lived stepping stone to GPT-5.4 and GPT-5.5, but the firmer signal is the usage number from OpenAI itself.