How do I actually use GPT-5.5 in my Codex? (SDK v0.130.0)
How do I force Codex to actually use the 5.5 model (or at least 5.4), without giving it a choice?
I've been using Claude for a long time and recently switched completely to Codex because Claude became too dumb even on the maximum tier. But I've encountered the fact that Codex chooses which model to work with on its own. I specify a particular model (for example 5.4) but over time I observe degradation in responses. I check with a standard method (tell me the year, without relying on environment and external sources) and if the agent starts forgetting context, writing bad code and making things up then the response year will always be 2024.
It's impossible to work with this. Claude is at least honest (either always good or always bad). How can you use a tool that works however it wants, and this isn't even tied to global time or anything like that.