The older large language models that have been widely relied upon by developers are about to face a "mandatory retirement" initiated by the platform. The Codex official team recently announced that starting from June 2, 2026, several older versions of models such as GPT-5.2 and GPT-5.3-Codex will be officially discontinued. This move by the platform aims to fully promote its latest cutting-edge flagship model, GPT-5.5, but this decision has sparked considerable controversy among developers.

Reactions to the Discontinuation of Core Old Models

In recent days, many developers have openly complained that the latest GPT-5.5 model exhibits obvious "dumbing down" behavior in practical applications, with a noticeable decline in logical processing capabilities and longer time required to handle issues. To maintain business stability, many developers had previously switched back to more mature versions such as GPT-5.3-Codex or GPT-5.4. Now, the decision by the Codex team to forcibly discontinue some old versions means that free users will no longer be able to switch, and will be forced to use the controversial GPT-5.5 model.

Paid Plans and Special Models Retained with Buffer Periods

To ease developers' dissatisfaction, the Codex announcement also outlined differentiated model retention policies for different groups. For subscription-paying users who log in through ChatGPT, although GPT-5.2/5.3 can no longer be used, the platform will still retain GPT-5.4 and GPT-5.5 for their selection. Meanwhile, commercial developers who call the model via API are temporarily not affected by this discontinuation policy. Additionally, the special model GPT-5.3-Codex-Spark, which is known for its extremely fast response speed, has also been included in the exemption list, allowing developers who have obtained permissions to continue using it normally in the future.