[Feature Request] Fast mode spend is invisible until it runs out: no line item, no in-session balance, no warning before exhaustion
Preflight
- [x] I searched existing issues. Closest are linked under Related. This is not about the billing shape (#92745 / #83302 / #87063). It is about fast mode spend being unobservable — before, during, and after it happens.
- [x] This is a single feature request.
What happened
I am on a Max plan and have been using fast mode. Mid-session, with plenty of plan usage left, fast mode shut itself off:
Fast mode disabled · usage credits exhausted. Fast mode was turned off. Add usage credits to turn it back on.
That notification was the first time I learned that fast mode draws on a separate balance at all. I did not know usage credits existed as a concept, let alone that the speed toggle was spending them.
The core problem: I still cannot determine what happened
After reading the docs and checking every surface available to me, I cannot answer basic questions about my own account:
- Were credits ever enabled? I do not know. If they were, fast mode has been spending real money and I never saw it. If they were not, then the toggle was reporting a balance I never opened as "exhausted." The notification does not distinguish these. #41605 reports the exhaustion wording appearing with credits never enabled and a $0.00 balance, so the message is not a reliable signal of which state you are in.
- What did fast mode cost me? Unanswerable by design. Per the docs:
> the Usage credits section shows how much you've spent in usage credits this month. That figure includes fast mode but doesn't break it out separately.
- How much was left before it ran out? Never shown. In Settings > Usage I found no mention of fast mode or of speed consuming anything, and nothing in the session surfaced a balance or a burn rate while fast mode was on.
The ↯ icon tells you fast mode is on. Nothing tells you it is costing anything, how much, or how much remains.
Why the one-time cost makes this worse
Documented behavior:
The first time you enable fast mode in a conversation, you pay the full fast mode uncached input token price for the entire conversation context.
I run very large, long-lived contexts. Every mid-conversation toggle re-bills the whole loaded context at $10/MTok before generating a token. That is plausibly the largest single line of fast-mode spend on my account, and it is invisible when incurred and aggregated away afterward.
Concrete ask
- Warn before exhaustion, not at it. A notification at a configurable threshold — a percentage of balance or a dollar amount remaining — so the shutoff is anticipated rather than discovered mid-work. This alone would have prevented the entire experience.
- Make the exhaustion notification diagnostic. State which situation the user is actually in: credits enabled and now at zero, versus credits never enabled. Include what the balance was. As written, the same sentence covers both and explains neither.
- Break fast mode out as its own line item in Settings > Usage. The docs currently document the absence of this as intended behavior; it should be the opposite. A user cannot evaluate whether a 2x-priced feature is worth it without knowing what it cost.
- Expose remaining credit balance in-session — status line and/or
/fastoutput — so cost is visible while being incurred, not only as a monthly aggregate on the web. (Adjacent to #88781.) - Disclose the dependency at the point of use. The first time a user enables fast mode, say plainly that it bills to usage credits and not to the plan, and show the approximate context size about to be re-billed at fast rates.
Related
- #92745 — fast mode billing shape on the desktop app; its ask #2 covers a warning at the toggle. This issue covers visibility during and after spend, which it does not. Open.
- #83302, #87063 — plan-included / plan-metered fast mode. Open.
- #41605 — exhaustion wording with credits never enabled. Closed as not planned. Cited here as evidence the message does not distinguish account states.
- #88781 — expose usage-credit state to the session. Open, directly adjacent to ask 4.
Environment
- Claude Code 2.1.226, Claude Desktop app (Code tab)
- Windows 11 Pro 10.0.26200
- Model: Opus 5
- Plan: Max subscription. Whether usage credits are or were enabled, I cannot determine from any surface available to me — which is the report.