github-copilot: surface & optionally pin context-pricing tier (default vs long_context) #38164

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Problem

GitHub Copilot bills GPT-5.x in two context tiers inside a single model: a default tier and a long_context tier that costs ~2× input / 1.5× output, switching automatically once the prompt crosses ~272K tokens. Copilot's own CLI lets you pick the tier per session. opencode shows neither the tier nor any control over it — so for models whose long window is their catalog default (gpt-5.5, gpt-5.6-luna/sol/terra at 1.05M context), a session silently grows past 272K into the 2× tier with no signal.

Evidence

From the live GET /models endpoint, e.g. gpt-5.6-sol.billing.token_prices (credits/1M):

(The default row matches Copilot CLI's "Default" pricing panel exactly.) The tier is an in-record billing field auto-selected by prompt size — not a separate model ID and not a request param.

This data already exists upstream — models.dev encodes it as cost.tiers (e.g. gpt-5.6-sol.toml has tier = { type = "context", size = 272_000 }). opencode parses it (packages/opencode/src/provider/provider.ts:1183) but uses it only for the passive cost estimate — never for window sizing or a visible tier signal. So this is an opencode feature gap, not a models.dev data bug.

Current workaround

Hand-set limit.input: 272000 per model in opencode.json (caps the prompt at the tier boundary so compaction keeps you on default). Works, but must be repeated for every model and every new model Copilot ships, and is invisible to anyone unaware the boundary exists.

Proposed (either is useful alone)

1. Opt-in tier pin — a provider/model setting (e.g. contextTier: "default" | "long") that sets the effective window to the model's cost.tiers boundary. Auto-generates today's manual override.
2. Tier indicator / warning — show the active tier and warn when a session crosses the boundary into the 2× tier.

Prior art

• Gemini: Context Tier Configuration, The 200K threshold price difference #592 — Gemini: Context Tier Configuration (200K threshold) — same concept, help-wanted, maintainer called it "a great optimization"; went stale/auto-closed, never built.
• Related context-limit confusion:
  • Discrepancies on context sizes using Opencode/Copilot CLI #17582
  • GPT-5.2 context limit for Github Copilot is too low #12247
  • Is there a way to increase Context limit for Github Copilot Provider? #5993
  • What is the specific context size of Github Copilot, is it only 128K? #16129