Reported by @night nova
Bug Report: GPT-5.2 Thinking redirects to GPT-4o-mini
`Steps to Reproduce`
Occasionally the model selector ignores the user-selected model, and responses are generated by GPT-4o-mini, although they still show in the web interface as having been generated by the originally selected model. This happened to me yesterday for a while but I am unable to reproduce it today.
`Expected Result`
Generate the response with the correct model, or maybe inform the user that the model has been switched due to server load/other causes
`Actual Result`
Response generated by 4o-mini without any visible indication (other than lack of thinking time and decreased response quality)
`Environment`
Chrome 143 on macOS (ChatGPT web interface)