[BUG] GPT 5.5 has a window context of only 400K. #237

It looks like the current behavior is expected for Codex subscriptions using a ChatGPT session.

When base_url is configured to use the ChatGPT Codex backend:

https://chatgpt.com/backend-api/codex

the actual context window is limited to 400K tokens, even for GPT-5.5 models.

Only requests authenticated with an API key via the official APIs currently get the 1M token context window:

Source:
https://codex.danielvaughan.com/2026/04/25/gpt-5-5-million-token-context-window-codex-cli-long-context-workflows/

So in practice:

• ChatGPT session + https://chatgpt.com/backend-api/codex → 400K
• API key + Responses API / Chat Completions API → 1M

Looks like this was already adjusted in this commit:
b5ad6d9

Looking into the context-window cap for GPT-5.5 (showing 400K vs the ~1M available via Codex OAuth). Note this builds on the gpt-5.5 model addition in #265 — the PR may depend on that landing first; I'll scope it to the context-window metadata.

GPT-5.5 is 1M only via the API; in Codex (Plus/Pro/Business/Enterprise/Edu/Go plans) it's 400K. Zoo Code's openai-codex provider is the Codex/OAuth path, so the current 400K is correct — raising it to 1M actually triggers "exceeds the context window of this model" errors (cf. openai/codex#19319, opencode#24171). So this is expected for the Codex provider rather than a bug — the 1M window is available via the API provider. 🦘

Thank you for the explanation.