Fast mode is the most expensive switch in Claude Code, and right now it is also the one with a deadline. On October 30, 2026 Anthropic deprecated fast mode for Opus 4.8, and on November 30, 2026 it is removed. If your sessions still run Opus 4.8 with fast mode on after that date, the requests fail instead of quietly falling back.
This post covers what fast mode costs, how it is billed on a subscription, how to turn it on and off, what to change before November 30, and why it adds up faster when you run several agents at once.
What fast mode is
Fast mode is not a separate model. According to the Claude Code fast mode docs, it runs Claude Opus with an API configuration that favours speed over cost: up to 2.5x faster output, with identical quality and capabilities. The API docs add that the gain is in output tokens per second, not in time to first token, so you notice it most on long answers and big edits.
It works on Opus 5.5, Opus 5 and Opus 4.8 only. Sonnet and Haiku do not have it. Fast mode on Opus 5.5 is generally available; on Opus 5 and Opus 4.8 it is a research preview.
What it costs
Fast mode is priced at twice the standard rate of each model. Prices per million tokens, from the Claude pricing page:
| Model | Standard input / output | Fast mode input / output |
|---|---|---|
| Opus 5.5 | $4 / $20 | $8 / $40 |
| Opus 5 | $5 / $25 | $10 / $50 |
| Opus 4.8 | $5 / $25 | $10 / $50 |
Two details matter more than the table suggests:
- Turning it on mid-conversation costs extra. The first time you enable fast mode in a conversation, the whole context is billed once as uncached input at the fast price. Fast and standard speed keep separate caches, so every later switch re-reads part or all of the history. On a 200,000-token conversation, that first switch on Opus 5.5 is 200,000 × $8 per million, or $1.60, before Claude writes a word. Enable it at the start of a session instead.
- Fast pricing is flat across the 1M context window. Prompt caching and data residency multipliers apply on top.
For how the standard plans and API rates compare overall, see our Claude Code pricing guide.
How it is billed on Pro, Max and Ultra
This is where most people get surprised. On subscription plans (Pro, Max, Ultra, Team and Enterprise), fast mode is available through usage credits only and is not included in the plan’s rate limits. The docs are explicit: fast mode draws directly from usage credits, even if you have usage left on your plan.
So a bigger plan does not make fast mode cheaper. If you are choosing between plans, our Claude Max vs Ultra comparison covers what the plan itself includes; fast mode sits on top of it.
What you need first:
- Pro, Max, Ultra: turn on usage credits in Settings > Usage on claude.ai, or run
/usage-creditsto open that page. Until then/fastreports “Fast mode requires usage credits”. - Team and Enterprise: fast mode is off by default. An Owner has to enable it under Organization settings > Claude Code, and a member with billing access has to turn on usage credits.
- Claude Console (API key): fast mode is billed per token with the rest of your API usage. On the free Evaluation plan it is unavailable.
In Claude Code, fast mode works only with the Anthropic API or a Claude subscription, not on Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry. To see what you spent, run /status to check your login method, then look at Settings > Usage (subscriptions, where fast mode is included in the usage-credits total) or the Console Usage and Cost pages grouped by Speed.
How to turn it on and off
- Toggle: run
/fast, press Space, then Enter. You see “Fast mode ON” and a↯icon next to the prompt. Run/fastagain to check or switch it off. - Settings:
"fastMode": truein your user settings file. Claude Code saves the/fasttoggle there too. - VS Code: the Toggle fast mode command, when the selected model supports it.
- Headless: with
-p, fast mode works only when it is in the launch settings, for exampleclaude -p --settings '{"fastMode": true}'(v2.1.205 or later). - Turn it off for good:
CLAUDE_CODE_DISABLE_FAST_MODE=1.
If your current model does not support fast mode, /fast switches you to Opus. Switching to a model without fast mode, such as Sonnet, turns it off, and switching back to a supported Opus turns it on again if your saved preference is on.
Fast mode also has its own rate limits. When you hit them, Claude Code falls back to standard speed, the ↯ icon turns gray, and fast mode comes back after the cooldown. If you run out of usage credits, requests are retried at standard speed and you see “Fast mode disabled · Out of usage credits”.
What to do before November 30, 2026
The Claude Code changelog entry for v2.1.316 (October 31, 2026) is short: fast mode on Opus 4.8 is deprecated as of October 30 and removed on November 30, 2026, and afterwards those requests “return an error and do not fall back to standard Opus 4.8.” This has happened before: fast mode for Opus 4.7 was deprecated on June 25, 2026 and removed on July 24.
- Check your model. Run
/statusor/modelin each project where you use fast mode. - To keep the speed: use
/modelto switch to Opus 5.5. It needs Claude Code v2.1.280 or later, fast mode stays on, and the per-token price drops from $10/$50 to $8/$40. Opus 5.5 is the fast mode default from v2.1.280. - To stay on Opus 4.8: run
/fast offand keep standard speed. - Check pinned models. If a new session still starts on Opus 4.8, a settings file, managed settings or an environment variable such as
ANTHROPIC_DEFAULT_OPUS_MODELis probably pinning it. - API users: after the removal, requests to
claude-opus-4-8withspeed: "fast"return an error. Opus 5.5 fast mode does not need thefast-mode-2026-02-01beta header and has a default fast mode rate limit in the Start, Build and Scale usage tiers.
What it means when you run several agents
Fast mode is designed around one person who wants faster answers in one session. With several agents running side by side, three things change:
- The setting follows you into every new session. By default, fast mode you turn on persists across sessions on a supported model. Open four Claude Code agents in four worktrees and all four can start in fast mode, each billing usage credits at twice the rate. The docs suggest
"fastModePerSessionOptIn": truefor exactly this case, “organizations where users run multiple concurrent sessions”: every session then starts with fast mode off until you run/fast. - Speed matters only where you are watching. Anthropic recommends fast mode for live debugging and rapid iteration, and standard mode for long autonomous tasks, batch work and CI. An agent working through a refactor while you review another one gains little from being 2.5x faster.
- Fast mode rate limits are separate. Several fast sessions share them, so with many agents you hit the cooldown sooner and fall back to standard speed anyway.
A practical setup: per-session opt-in on, fast mode only in the one agent you are actively steering, standard speed (or Sonnet) everywhere else.
Where AgentVera fits
We build AgentVera, a desktop app for macOS, Windows and Linux that runs Claude Code, Codex and 16 other coding agents side by side, each in a real terminal with your own install and login. It does not change how Anthropic bills fast mode, but it helps you see and limit what parallel agents spend:
- Token budgets: caps per agent, per turn and per day, with a warning at 80%. At the cap the agent stops but keeps its conversation.
- Automatic model: rates each task simple, normal or hard and moves Claude to Haiku, Sonnet or Opus, so Opus is kept for the hard work.
- Usage at a glance: the status bar shows each Claude and Codex account’s 5-hour and weekly usage. On Pro, automatic account switching moves an agent to the next account when one hits its limit.
The free plan includes up to 18 agents, MCP, Git, Docker, SSH, code review and token savings. Pro is $9.99 a month with a 7-day trial. For a wider look at tools in this space, see the best apps to run Claude Code and Codex in parallel.
Wrapping up
Fast mode buys up to 2.5x faster Opus output for twice the token price, paid from usage credits even on Max and Ultra. If you use it on Opus 4.8, switch to Opus 5.5 or run /fast off before November 30, 2026. And if you run several agents, turn on fastModePerSessionOptIn so speed is something you choose per session, not a default every agent inherits.
Sources: Speed up responses with fast mode, Claude Code changelog, Fast mode on the Claude API and Claude pricing.