GitHub Copilot Business gpt-5.5 context differs from models.dev

Description

I noticed a context limit mismatch for github-copilot/gpt-5.5.

The local models.dev-derived cache used by OpenCode shows:

github-copilot/gpt-5.5
limit.context: 400000
limit.input: 272000
limit.output: 128000

But the live GitHub Copilot models endpoint for my account reports:

GET https://api.githubcopilot.com/models
model: gpt-5.5
max_context_window_tokens: 1050000
max_prompt_tokens: 922000
max_output_tokens: 128000

Account / plan context

This is from a GitHub Copilot Business account, not a personal Copilot account.

So this may be plan/account dependent rather than a universal Copilot limit.

Why I noticed

OpenCode TUI showed the session at around 10% used, which matched the live 1050000 context value.

However, reading the local models cache made it look like the same session was much fuller because models.json listed 400000 for the same model.

Related OpenCode issue

I also filed the OpenCode-side observability issue here:

https://github.com/anomalyco/opencode/issues/31064

Start by comparing the models.dev-derived cache entry with the live GET https://api.githubcopilot.com/models response for gpt-5.5, keeping the GitHub Copilot Business account context in mind. Determine whether the data should represent plan-specific limits or a documented canonical value, then verify that the resulting model metadata no longer creates the reported context mismatch.