Max plan 1M context window not used — compaction triggers at 200K

Description

On the Max subscription plan, Claude Code compacts conversation context at ~200K tokens instead of using the advertised 1M token context window for Opus 4.6.

Environment

• Plan: Max (1M context window)
• Model: Claude Opus 4.6 (claude-opus-4-6)
• Claude Code: CLI (latest)
• OS: macOS (Darwin 24.6.0)

Expected behavior

With the Max plan, the context window should be 1M tokens. Compaction should trigger near that limit, not at 200K.

Actual behavior

Compaction triggers at approximately 200K tokens — the default context window size for Opus 4.6 on non-Max plans. The status line also reports (200K) as the total context capacity, confirming that Claude Code does not recognize the extended 1M limit.

Steps to reproduce

1. Subscribe to Max plan
2. Start a Claude Code session with opus model
3. Work on a large task that fills context past ~180-200K tokens
4. Observe automatic compaction at ~200K instead of continuing to 1M

Additional context

• No compaction-related settings exist in ~/.claude/settings.json or project settings
• No environment variables (CLAUDE_AUTOCOMPACT_*) are set
• The status line (read-only display) shows (200K) as total capacity, suggesting Claude Code itself reports 200K regardless of subscription tier
• This means Max plan users are paying for 1M context but only getting 200K in Claude Code

Workaround

None known. There is no user-facing configuration to set the context window size or compaction threshold.