GPT-5.6 Sol context window remains limited to 272k despite 1M configuration

What version of Codex CLI is running?

26.819.11345

What subscription do you have?

pro

Which model were you using?

gpt-5.6-sol xhigh

What platform is your computer?

macOS

What terminal emulator and version are you using (if applicable)?

Terminal.app

Codex doctor report

What issue are you seeing?

Description

GPT-5.6 Sol is still limited to a 272k context window on my ChatGPT Pro account, even when I explicitly configure Codex to use the 1M context window.

According to the documented configuration, I tried both config.toml and CLI overrides, but neither changes the effective context window.

Configuration

In ~/.codex/config.toml:

model = "gpt-5.6-sol"
model_context_window = 1000000
model_auto_compact_token_limit = 900000

I also tested a fresh CLI session with:

codex -m gpt-5.6-sol
-c model_context_window=1000000
-c model_auto_compact_token_limit=900000
Actual behavior

After starting a new session and sending messages, the backend/model metadata still reports GPT-5.6 Sol with a maximum/effective context window of approximately:

272000

The requested model_context_window=1000000 does not take effect.

The same issue occurs after:

restarting Codex
starting a completely new conversation/session
logging out and logging back in
refreshing/removing the local model cache
using the configuration file
using CLI -c overrides directly

The context window remains limited to 272k.

What steps can reproduce the bug?

Uploaded thread: 01a0296b-839a-7c40-b643-59914df5741c

What is the expected behavior?

Expected behavior

GPT-5.6 Sol supports a much larger context window, and Codex has documented a configuration for enabling the 1M-token context window:

model_context_window = 1000000
model_auto_compact_token_limit = 900000

I therefore expect Codex to honor the larger context configuration instead of limiting the model to 272k.

Additional information

Possible cause

This appears to be related to server-delivered model metadata or an account-level entitlement/configuration issue.

It looks like the server returning a max_context_window of 272000 for GPT-5.6 Sol, which causes the locally configured 1M context window to be clamped back to 272k.

“slug: gpt-5.6-sol
context_window: 272000
max_context_window: 272000
effective_context_window_percent: 95”

Could you please check whether the GPT-5.6 Sol long-context capability is being correctly enabled for my ChatGPT Pro account?

Start by reproducing the issue with the config.toml values and the CLI -c overrides, then inspect the reported model metadata showing context_window and max_context_window at 272000. Determine whether the server-delivered metadata or account entitlement clamps the local 1M setting. Done means identifying the source of the clamp and confirming that the configured context window is honored or documenting the external limitation.