stuck at "100% context left" for local models

What version of Codex is running?

0.50.0

What subscription do you have?

local model

Which model were you using?

gpt-oss-20b

What platform is your computer?

Linux 6.14.0-33-generic x86_64 x86_64

What issue are you seeing?

Codex doesn't seem to update how much context is left, and therefore never auto-compacts. The status always indicates "100% context left", no matter how much context I use.

What steps can reproduce the bug?

My config is simple:

[model_providers.lms]
name = "LM Studio"
base_url = "http://localhost:1234/v1"

[profiles.lmstudio]
model_provider = "lms"
api_key = "not-needed"
model = "openai/gpt-oss-20b"
model_context_window = 131072

approval_policy = "on-failure"

and then I run:

codex --profile lmstudio "hi, write a 2 page story"

At the end of the command, there's still "100% context left" in the status.

What is the expected behavior?

After 2 pages of story, I expect the context to drop below 100%

Additional information

I also tried previous versions of Codex and it never seemed to work. Context seems to be properly updated with other CLI tools such as gemini afair