Claude Code fast mode is a speed setting for Claude Opus: run /fast and Opus answers up to 2.5x faster, with the same quality, at a higher price per token. It suits live debugging and quick back-and-forth edits, and you turn it off again when cost matters more than waiting.
This guide covers how to toggle it, what it costs as of October 2026, who can use it and when it pays off. Everything here comes from the official fast mode documentation, which labels the feature a research preview, so pricing and availability can change.
What is fast mode in Claude Code?
Fast mode is not a separate model. The docs describe it as Claude Opus "with a different API configuration that prioritizes speed over cost efficiency." You get identical capabilities and output quality; the only things that change are latency and price.
Supported models are Opus 5.5, Opus 5 and Opus 4.8. Sonnet and Haiku do not have a fast mode. Opus 4.7 lost fast mode support on July 24, 2026, so switching to it turns fast mode off. In Claude Code v2.1.280 and later, Opus 5.5 is the default model when you turn fast mode on.
Fast mode is also different from lowering the effort level. The two settings speed things up in different ways:
| Setting | What changes |
|---|---|
| Fast mode | Same model quality, lower latency, higher cost |
| Lower effort level | Less thinking time, faster responses, possibly lower quality on complex tasks |
You can combine them. Fast mode with a lower effort level is the quickest setup for simple, well-defined tasks.
How do you turn fast mode on and off?
In the CLI you have three options:
- Run
/fast, press Space to toggle, then Enter to confirm. - Press
Option+Oon macOS orAlt+Oon Windows and Linux, a shortcut listed in the interactive mode reference. - Set
"fastMode": truein your user settings file.
{
"fastMode": true
}
When fast mode turns on you see "Fast mode ON", and a small ↯ icon appears next to the prompt. Run /fast again at any time to check its state. If your current model does not support fast mode, turning it on switches you to Opus. Turning it off leaves you on Opus; use /model to change models.
By default, fast mode you turn on in an interactive session carries over to new sessions. You can run /fast while Claude is mid-task: the running turn finishes at its original speed and the change applies from your next turn.
Other places it works:
- VS Code extension: a Toggle fast mode command appears when the selected model supports it.
- Cloud sessions: type
/fast onin the session (Claude Code v2.1.271 or later). It lasts for that session only. On claude.ai/code you can also flip it from the model menu. - Non-interactive mode:
/fastonly works in a-psession launched with fast mode in its settings, such asclaude -p --settings '{"fastMode": true}', and needs v2.1.205 or later.
How much does Claude Code fast mode cost?
As of October 2026, the fast mode docs list these prices per million tokens:
| Model | Fast mode input | Fast mode output | Standard input | Standard output |
|---|---|---|---|---|
| Opus 5.5 | $8 | $40 | $4 | $20 |
| Opus 5 | $10 | $50 | $5 | $25 |
| Opus 4.8 | $10 | $50 | $5 | $25 |
Standard rates are from the Claude API pricing page, also as of October 2026. In other words, fast mode costs twice the standard Opus rate, and that price is flat across the full 1M token context window.
One detail matters for your bill. The first time you turn fast mode on in a conversation, you pay the full fast mode uncached input price for the entire conversation context so far. The deeper you are into a session, the bigger that one-time charge. It applies once per conversation, so toggling off and on later does not repeat it. The practical rule from the docs: turn fast mode on at the start of a session rather than halfway through.
Who can use fast mode?
Fast mode works on Claude subscription plans (Pro, Max, Team and Enterprise) and with Claude Console API accounts. The Claude Code docs say it is not available on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or Claude Platform on AWS. The Claude API pricing page is broader for direct API use: it lists fast mode on the Claude API and, for Opus 5.5 only, on Claude Models on Google Cloud (Anthropic managed) and Claude in Microsoft Foundry for Azure-hosted deployments. For Claude Code, follow the Claude Code docs.
The billing rules differ by account type:
- Pro and Max: fast mode is paid from usage credits only, even if you still have plan usage left. Turn on usage credits in Settings > Usage on claude.ai, or run
/usage-credits. Until you do,/fastreports "Fast mode requires usage credits". - Team and Enterprise: fast mode is off by default. An Owner must enable it at Organization settings > Claude Code, and someone with billing access must turn on usage credits for the organization.
- Claude Console: your organization pays per token with the rest of its API usage. Fast mode is not available on the free Evaluation plan, and the organization needs fast mode access provisioned (through an account manager or the waitlist) before requests succeed.
Not sure how you are signed in? Run /status. A Login method row such as "Claude Max account" means a subscription; an API key row means Console billing.
When is fast mode worth it?
The docs recommend fast mode for interactive work where waiting costs you more than tokens do:
- rapid iteration on code changes
- live debugging sessions
- time-sensitive work with tight deadlines
Standard speed is the better choice for:
- long autonomous tasks where nobody is watching the screen
- batch processing and CI/CD pipelines
- cost-sensitive workloads
A simple way to decide: if you are sitting there reading each response as it arrives, faster responses save your time. If Claude is working through a long task in the background or in headless mode, speed buys you little and doubles the bill. For a broader look at spending, see the Claude Code cost guide.
What happens when you hit a fast mode rate limit?
Fast mode has its own rate limit pool, separate from standard Opus, shared across all supported Opus models. When you hit it, Claude Code falls back to standard speed automatically, the ↯ icon turns gray during the cooldown, and you keep working at standard speed and price. When the cooldown ends, fast mode turns back on by itself. Run /fast if you would rather switch it off than wait.
If you run out of usage credits mid-session in an interactive session, Claude Code retries each request at standard speed, shows "Fast mode disabled · usage credits exhausted", and turns fast mode off for the rest of that session. Your saved preference stays on.
How can teams control fast mode spend?
Admins have a few levers, all documented on the fast mode page:
- Per-session opt-in: set
fastModePerSessionOptIntotruein a settings file so every session starts with fast mode off. This helps when people run several sessions at once. - Turn it off entirely: set
fastMode: falsein managed settings, or set the environment variableCLAUDE_CODE_DISABLE_FAST_MODE=1. - Restrict models: an
availableModelsallowlist that excludes the fast mode Opus model blocks turning it on.
{
"fastModePerSessionOptIn": true
}
Spend shows up in different places. Pro and Max users see it inside the Usage credits total on claude.ai (not broken out separately). Team and Enterprise members can run /usage. Console admins can group the Usage and Cost pages by Speed (Research Preview).
Why does fast mode say it is unavailable behind a gateway?
Before offering fast mode, Claude Code checks availability with a direct request to api.anthropic.com. That check ignores ANTHROPIC_BASE_URL, so on a network that routes traffic through an LLM gateway and blocks direct access, /fast reports "Fast mode unavailable due to network connectivity issues" even though normal requests work.
The fix is to allowlist api.anthropic.com, or set one of two variables. CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS=1 treats a failed check as available. CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK=1 skips the check entirely, which is the one to use when a proxy intercepts the request or when you authenticate with ANTHROPIC_AUTH_TOKEN alone. Neither variable overrides an organization that has fast mode disabled: the API still rejects those requests.
Get more out of every fast session
Fast mode pays off most when your prompts are already sharp, so fewer turns go to clarifying what you meant. Kipt keeps the prompts you reuse as Markdown in a private GitHub repo you own, with {{name}} blanks to fill in per task and Cmd+K search to find one while a Claude Code session is open.