Show context window size (tokens) in /model command alongside premium request multiplier

[svg153]

Describe the feature or problem you'd like to solve

The /model command shows the premium request multiplier for each model but not the context window size (max tokens). There is no way to know, from within the CLI, how many tokens a model actually supports in Copilot.

Proposed solution

Add the context window size (in tokens) to the output of /model, alongside the existing premium request multiplier.

Current /model output:

  claude-sonnet-4.6      ×1  (default)
  claude-opus-4.6        ×3

Proposed /model output:

  claude-sonnet-4.6      ×1    160k ctx  (default)
  claude-opus-4.6        ×3    160k ctx

The context window values should reflect what Copilot actually makes available (i.e. the max_prompt_tokens configured in its routing layer), not just the provider's theoretical max. This distinction matters because the effective context window Copilot exposes may differ from the provider's advertised limits — and today users have no visibility into this.

Example prompts or workflows

1. Choosing between models for a long session: A developer about to start a large refactor touching 30+ files runs /model to pick the model with the largest effective context window, not just the cheapest in premium requests.

2. Debugging degraded responses: A user notices the model is repeating tool calls and re-reading already-processed files. Knowing they're at 160k/200k tokens (vs just "78%") helps correlate the degradation with the known "lost in the middle" effect at high context fill.

3. Cross-client comparison: A user switching between VSCode Copilot, Copilot CLI, and Claude Code wants to know if they get the same effective context window across clients. Today this is completely opaque.

Additional context

• The data is likely already available in the API response from api.githubcopilot.com/models (the capabilities.limits field), so this should be low-effort to surface.
• Related issues: #1688 and #1761 both request configurable compaction thresholds — knowing the absolute context window size would make those features even more useful.

[shantanugoel]

You can see it with /context They seem to have increased the input context size by 32k for opus 4.6/gpt 5.4 probably this week.