[Feature Request] Fast mode in the desktop app (Code tab): meter it against plan usage at a multiplier, like Codex, instead of a separate usage-credits wallet

### Preflight

- [x] I searched existing issues. The closest ones are linked under [Related](#related); this one is scoped to the desktop app and to the billing *shape*, not to a bug.
- [x] This is a single feature request.
- [x] I am on the latest version available on this surface (Claude Code 2.1.263, Claude Desktop app — Code tab, macOS).

## What I want

Be able to turn fast mode on in the **Claude Code desktop app (Code tab)** and have it billed against the subscription I already pay for, at a published multiplier — the way Codex bills fast mode as a **2.5x multiplier on the same balance** — instead of it requiring a separate usage-credits wallet to exist at all.

## Why the current shape makes it unusable, and why it is not a "2x" decision

The docs are explicit for subscription plans:

> fast mode is available via usage credits only and not included in the subscription rate limits

> Fast mode usage draws directly from usage credits, even if you have remaining usage on your plan

Put next to the rate card, that wording hides how large the decision actually is:

| | Input / MTok | Output / MTok | Marginal cost to a subscriber with plan headroom |
|---|---|---|---|
| Opus 5, standard | $5 | $25 | **$0** (covered by the plan) |
| Opus 5, fast | $10 | $50 | **full price, from the first token** |

The table says 2x. The decision a subscriber actually faces is **$0 → paid**. Toggling `/fast` does not make my session 2x more expensive; it moves me out of the subscription billing model and into the pay-per-use one, mid-flight. That is a different kind of choice than "pay a bit more for speed", and it is the reason the feature stays off on my account even in sessions where I would happily pay for it.

Codex's shape avoids exactly this. Fast mode there is a multiplier (2.5x Standard; 2x on GPT-5.4) applied to the **same** credit pool the user is already spending. One wallet, one decision, a legible price for speed. On Claude Code it is two wallets — and the second one has to be opened before the feature exists at all:

> Fast mode requires usage credits · /usage-credits to turn them on

So the funnel for a subscriber is: read the docs, discover the feature is gated behind a second billing channel, and stop. Not "I evaluated fast mode and chose standard speed."

### What I would spend it on

Interactive inner-loop work — Java/Spring, large context, live debugging — where latency *is* the product. Rough shape of a typical session of mine: ~200k output tokens over ~3h. At ~2.5x generation speed that is on the order of 20–30 minutes of wall clock returned, once tool calls, builds and test runs are accounted for. I would pay a plan multiplier for that without thinking about it. What I will not do is open a second pay-per-use billing channel to find out whether it is worth it.

## Concrete ask

1. **Let plan usage cover fast mode at a published token-equivalent multiplier.** Fast tokens draw N× against the plan's limits rather than requiring a separate balance. Heavy fast-mode use then drains the plan faster, which is honest pricing for speed, instead of being all-or-nothing for subscribers. This is the same shape asked for in #83302 and #87063; I am restating it because the desktop app is where I run into it, and because the Codex comparison gives it a concrete, shipped precedent.

2. **Surface the consequence at the toggle in the Code tab, not only in docs.** Whether this session will draw plan usage or credits, the current credit balance, and — most importantly — the one-time cost of enabling mid-conversation:

> 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.

That is invisible today and it is the most expensive surprise in the feature. On a large loaded context it is real money spent before a single token is generated. A one-line warning at the toggle ("enabling now re-bills ~X tokens of context at fast rates; enabling at session start avoids this") would change behaviour.

3. **Document the desktop app as a supported surface on the fast mode page.** [The page](https://code.claude.com/docs/en/fast-mode) documents the CLI (`/fast`, `"fastMode": true`) and the VS Code extension (a **Toggle fast mode** command) and says nothing about the desktop app. From the docs alone there is no way to tell whether the Code tab supports fast mode, follows the `fastMode` setting, or honours `fastModePerSessionOptIn`.

## One adjacent note

Whatever the billing shape ends up being: fast mode only earns its price on turns a human is watching. Background subagents, Agent Team teammates, background tasks and scheduled runs get nothing from faster streaming — nobody is waiting on the tokens. If fast mode becomes plan-metered, having it default to foreground turns only (or be configurable that way) would keep it from quietly draining plan capacity on work nobody is watching.

## Related

- #83302 — same core billing ask (a plan-included or plan-metered fast mode), CLI framing. Open.
- #87063 — "Enable Fast Mode for Max 20 accounts without additional credits". Open.
- #86250 — `"fastMode": true` not applied at session start. Open.
- #90925 — `modelPerSessionOptIn` mirroring `fastModePerSessionOptIn`. Open.
- #82018 — desktop Code tab reported `Fast` in the status bar while every request ran standard. Closed as completed; this request is about the billing shape, not that bug.
- #64083 — `/fast` had no observable effect in the desktop app. Closed as not planned.
- #56926 — make `/fast` available without extra-usage enabled. Closed as not planned; reopening the discussion because fast mode is a research preview whose "availability and pricing are subject to change", and Codex has since shipped the one-wallet multiplier shape.

## Environment

- Claude Code 2.1.263, Claude Desktop app (Code tab)
- macOS, Darwin 25.6.0
- Model: Opus 5
- Plan: subscription (usage credits not enabled)

## Sources

- [Fast mode — Claude Code docs](https://code.claude.com/docs/en/fast-mode)
- [Pricing — Claude Platform docs](https://platform.claude.com/docs/en/about-claude/pricing)
- [Speed / fast mode — Codex docs](https://developers.openai.com/codex/speed)
- [Codex rate card — OpenAI Help Center](https://help.openai.com/en/articles/20001106)

Start with the Fast mode documentation and the linked issues, especially #83302, #87063, #86250, and #90925, then inspect the desktop app's Code tab behavior. Done means the billing model, toggle disclosure, and desktop-app support are defined and implemented across the relevant product surfaces; the payload names no source files or tests.

Written by the indexing model from the issue text.