Max plan: Claude Code may compress context at 200K instead of using extended context window

Plan: Max 20x
Model: claude-opus-4-6
Version: 2.1.153

Claude Code triggers context compression ("Transmuting") well before what my Max plan should allow. I suspect the client tracks the context window against the model's base 200K limit and doesn't account for the extended context available on Max plans.

Evidence:

• Session hits "100% context used" and starts compressing during moderate-length conversations
• ~82K tokens of system prompt (startup hooks, CLAUDE.md chain, 9 MCP servers) means only ~118K remains for actual conversation before compression
• On Max 20x, I'd expect access to the full context window, not early compression at 200K

Expected behavior: Claude Code should query the plan's actual context limit from the API and use that as the compression threshold, not a hardcoded model default.

Requested info: Does Max 20x extend the context window for Opus 4.6 beyond 200K? If yes, does Claude Code honor that?