Increase Codex CLI context window to match ~400k model capacity

What feature would you like to see?

Expose the full context window of the underlying gpt-5.2-codex model (~400k tokens) in the Codex CLI, or provide a config/flag to enable higher limits.

Additional information

Right now the Codex CLI seems capped at ~258k tokens, while the model itself supports ~400k. Not sure if this limitation comes from the CLI layer, the client-side streaming protocol, memory constraints, or a temporary safety limit. Would love clarification on why the CLI exposes only ~258k, and whether matching the full ~400k context is planned or technically feasible.

For real-world codebases (multi-repo, long-range refactoring, persistent editing), the extra context would make a big difference.