I’m wondering if gpt-5.2-chat-latest is the same as gpt-5.2 with reasoning set to none, or if there is a difference between these two models in the way they behave?
1 Like
gpt-5.2-chat-latest is a bit different than gpt-5.1-chat-latest: it is always in “reasoning mode”, not accepting sampling parameters, and cannot accept any reasoning value other than “medium” (or null) which is a default that you can pass to non-reasoning models on Responses also, so likely used here not for obtaining that value, but also as a default setting.
However, doesn’t seem to bill for reasoning unless you prompt it into reasoning with internal techniques - to prove it is capable, and that it is likely doing some initial judgements on your input:
| input tokens: 106 | output tokens: 493 |
|---|---|
| uncached: 106 | non-reasoning: 365 |
| cached: 0 | reasoning: 128 |
The output is peppered with emoji, so likely runs a different “system prompt injection” than the API models. It is the “version being run on ChatGPT”.