What is the type of issue?
Documentation is incorrect
What is the issue?
As clearly explained by Tibo, GPT-5.6 Sol within Codex supports a 1M context window:
https://x.com/thsottiaux/status/2089082893804896524
Here is how to enable a 1M-token context window in Codex for GPT-5.6 Sol.
Even though we have tuned the context limit in Codex to be set optimally when it comes to performance and cost, this is a common ask, so here it is documented.
...
model_context_window = 1000000
model_auto_compact_token_limit = 900000
This runs directly contrary to the written changes made in the Codex CLI 0.144.6 changelog where it explicitly states:
Refreshed bundled instructions for GPT-5.6 Sol, Terra, and Luna, and corrected their context windows to 272,000 tokens. (#33972, #34009)
Tibo's suggestion of model_context_window USED to work before that patch, but as it stands what Tibo suggested is simply not true and does not work in the Codex App or CLI, making this change does not increase the context window. If you boot the CLI with that and run /status it reports 1M, but after your first message it reports the real capped limit.
This can be replicated with the codex CLI via codex debug models --bundled reporting {"slug":"gpt-5.6-sol","context_window":272000,"max_context_window":272000}.
There is clearly a gap in understanding between what OpenAI believe can be accessed, versus what is publicly documented and available. Please help us understand.
Where did you find it?
No response
What is the type of issue?
Documentation is incorrect
What is the issue?
As clearly explained by Tibo, GPT-5.6 Sol within Codex supports a 1M context window:
https://x.com/thsottiaux/status/2089082893804896524
This runs directly contrary to the written changes made in the Codex CLI 0.144.6 changelog where it explicitly states:
Tibo's suggestion of model_context_window USED to work before that patch, but as it stands what Tibo suggested is simply not true and does not work in the Codex App or CLI, making this change does not increase the context window. If you boot the CLI with that and run /status it reports 1M, but after your first message it reports the real capped limit.
This can be replicated with the codex CLI via
codex debug models --bundledreporting{"slug":"gpt-5.6-sol","context_window":272000,"max_context_window":272000}.There is clearly a gap in understanding between what OpenAI believe can be accessed, versus what is publicly documented and available. Please help us understand.
Where did you find it?
No response