[1m] model variants in /model picker carry premium billing with no warning at selection or runtime

Selecting a [1m] model variant (e.g. claude-fable-5[1m]) from the /model picker persists it as default with no mention that requests over 200k input tokens bill at long-context premium rates and consume subscription usage substantially faster. There is also no runtime indicator when a session or subagent crosses 200k context.

I ran this way unknowingly from May 14 to June 11 — a transcript audit shows 300–4,200 premium-context requests per day, with subagents iterating for hours at ~280k context. Burned through daily/weekly limits with no explanation until I audited transcripts manually.

Ask:

1. A cost note on 1M variants in the /model picker
2. A visible indicator (statusline / /usage) when context exceeds 200k
3. /usage breaking out long-context consumption separately

Environment: Claude Code v2.1.173, Windows 11, Max plan.