Fast mode toggle has no billing warning for Max subscribers

[mktkr]

Summary

The /fast slash command in Claude Code can be toggled freely from any
subscription tier, but fast mode is actually billed at premium API rates
($30 input / $150 output per MTok — ~6x standard Opus rates) and does not
draw from a Max plan's included usage allocation.

Problem

A Max Pro subscriber can run /fast, see "Fast mode ON," and start
incurring separate pay-as-you-go charges with no in-CLI indication that
they've left their plan's billing umbrella. The behavior is documented on
a separate API docs page, but nothing in Claude Code itself signals it.

Suggested fixes (any one would help)

1. Gate the toggle behind a one-time confirmation dialog the first time
   it's enabled on a subscription account: "Fast mode bills at $30/$150
   per MTok outside your Max plan. Continue?"
2. Show a persistent indicator in the status line while fast mode is on,
   noting it's billed separately.
3. Disable /fast entirely for accounts without API billing configured,
   with a tooltip explaining why.

Repro

1. Sign in to Claude Code on a Max Pro subscription
2. Run /fast
3. Observe: toggles ON with no billing warning

Environment

• Claude Code CLI on macOS (Darwin 25.1.0)
• Subscription: Claude Max Pro

[afram123]

+1, and the mid-conversation toggle gotcha makes the missing warning even more punishing than the issue body suggests.

Per the official docs, enabling /fast partway through a session re-bills the entire prior context at fast-mode uncached input rates ($30/MTok), not just tokens going forward. So a Max subscriber who flips /fast after an hour of conversation isn't just "starting to incur extra-usage charges" — they're retroactively paying premium rates on every byte of context already accumulated in that session, with no indication this is about to happen.

Option 1 (one-time confirmation dialog) should be the floor, and it should specifically call out the uncached-prior-context billing, not just the per-token rate. The dialog text in the issue body undersells the actual cost surprise.

Related policy discussion in #56926 — the cleaner fix is to make fast mode available within Max plan limits in the first place, which sidesteps the warning-design problem entirely.

[github-actions]

Closing for now — inactive for too long. Please open a new issue if this is still relevant.