Claude Code fast mode not working

Start with what you are looking at

Run /fast with no argument before changing anything. Anthropic's page says the command reports whether fast mode is currently on, so a refusal you can read points at its own cause. Whatever comes back, one of the five entries below fits.

• There is no toggle, or fast mode will not turn on
• "Fast mode has been disabled by your organization"
• "Fast mode is currently unavailable", or a network message
• "Fast mode requires usage credits"
• It turns on, and you want to know what it costs

What fast mode changes, and what it does not

What it is. Anthropic describes fast mode as a high-speed configuration for Claude Opus, making the model up to 2.5 times faster at a higher cost per token. It is not a different model: the same Opus runs with a different API configuration that prioritises speed over cost efficiency, and Anthropic states that quality and capabilities are identical.

It is supported on Opus 5.5, Opus 5 and Opus 4.8. It is not available on Sonnet, Haiku or any other model. Opus 4.7 does not support it, so switching a session to Opus 4.7 turns fast mode off. Anthropic deprecated fast mode on Opus 4.7 on June 25, 2026 and removed it on July 24, 2026.

A variable that sounds related is not. ANTHROPIC_SMALL_FAST_MODEL names the cheap model behind background work, and Anthropic's model configuration page marks it deprecated in favour of ANTHROPIC_DEFAULT_HAIKU_MODEL. It has no connection to the /fast toggle, and setting it changes nothing about speed on Opus.

Sources: Anthropic, speed up responses with fast mode and Anthropic, model configuration.

There is no toggle, or fast mode will not turn on

What it means. Nothing is broken. Fast mode is offered only when the session could use it, so an absent switch is a statement about the account, the model or the build. Work down these in order, because each one is cheaper to check than the next.

Where the switch is drawn, and where it is not

In the terminal, a non-Opus model does not cost you the command. Anthropic says that turning fast mode on when the current model does not support it also switches the session to Opus, and that Claude Code uses the new model from its next request in that turn.

The graphical surfaces are where the switch can be absent, and Anthropic documents the condition for the two it names. The VS Code extension offers its Toggle fast mode command only when the selected model supports fast mode. In the browser at claude.ai/code, the model menu on the message box shows the switch when the plan includes fast mode and the selected model supports it. If you are looking at either surface on Sonnet or Haiku, that is the documented reason, and choosing an Opus model from the model menu brings the switch back. One case is not documented and appears in the issue tracker instead: a surface can be missing the switch for a model that does support fast mode, because the catalog that surface reads has no fast mode entry for that model. The section on the issue tracker below has the example.

The provider does not offer fast mode at all

Anthropic lists fast mode as available through the Anthropic Console API and for Claude subscription plans, and not on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or Claude Platform on AWS. A session running through any of those four cannot turn it on, and no setting changes that. This is worth checking before anything else if a team standardised on one of those platforms, because it explains a whole fleet with no toggle.

An organization switched it off

Fast mode is disabled by default for Team and Enterprise organizations, so an Owner has to enable it before anyone can use it. Anthropic names four organization settings that block the toggle, and they arrive as messages rather than as a missing toggle, so the section below covers them.

An environment variable turns it off

Anthropic documents CLAUDE_CODE_DISABLE_FAST_MODE=1 as a way to disable fast mode entirely. It is worth searching your shell profile, your launcher script and any managed environment for it, because a value set there follows every session that inherits it.

The availability check failed

Before offering fast mode, Claude Code asks Anthropic whether the organization has it. That request goes directly to api.anthropic.com rather than through ANTHROPIC_BASE_URL, so a network that blocks direct egress can fail the check while ordinary inference keeps working. The check does use a configured HTTP proxy, so it fails only where api.anthropic.com is unreachable even through the proxy. A check that succeeded once is cached, which is why a blocked network mostly affects new installations.

You are in a non-interactive run

With the -p flag, /fast works only in a session started with fast mode already in its --settings value, as in claude -p --settings '{"fastMode": true}'. In that form the toggle applies to that session and is not saved as a default, and it needs Claude Code 2.1.205 or later. Elsewhere in non-interactive mode the command reports that fast mode is not available.

In Anthropic's cloud sessions, check your version

Fast mode works in Claude Code's cloud sessions when the account has it, whether the session runs on Anthropic's infrastructure or a self-hosted runner, and it needs Claude Code 2.1.271 or later in that session's environment. The form there is /fast on, it lasts for that session only, and it is not saved as a default.

How to turn it on once it is available

1. Run /fast, press Space to switch it on or off, then Enter to confirm. The alternative is "fastMode": true in your user settings file.
2. Look for the confirmation. Anthropic's page says fast mode shows Fast mode ON and puts a small arrow icon beside the prompt while it is active.
3. Run /fast again at any time to read whether it is on or off.
4. Know that turning it off leaves you on Opus. Use /model if you want a different model as well.

Two defaults are worth knowing because they make fast mode look persistent when you expected it not to be, or the reverse. Fast mode you turn on in an interactive session persists across sessions unless fastModePerSessionOptIn is set, which starts each session with fast mode off. And the model it defaults to follows your build: Opus 5.5 from 2.1.280, Opus 5 from 2.1.219, and Opus 4.8 before that, from 2.1.154.

Sources: Anthropic, fast mode, its requirements section and its proxy section.

"Fast mode has been disabled by your organization"

Fast mode has been disabled by your organization.

is not in your organization's allowed models

What it means. Anthropic documents four organization settings that each produce a refusal when you turn fast mode on, and three of them print that first sentence. Which one applies decides who can fix it. The first needs an Owner, the second needs whoever deploys managed settings, the third is a choice about cost control, and the fourth is a model allowlist.

The four causes

• Fast mode was never enabled. It is off by default for Team and Enterprise. An Owner turns it on for the organization, and until then every member sees the refusal.
• Managed settings set fastMode: false. The same sentence appears, and only whoever controls the managed settings file can lift it.
• Managed settings set fastModePerSessionOptIn: true. This refuses /fast on in every context except an interactive terminal session, including non-interactive runs, the VS Code extension and Anthropic's cloud sessions.
• The allowlist excludes the fast mode model. If the organization's availableModels list leaves out the Opus model fast mode would use, turning it on is refused with the second sentence above. There is one exception worth reading twice: in a session already running on an allowed Opus model that supports fast mode, /fast switches fast mode on for the current model without changing models.

Where an Owner enables it

On Claude for Teams or Enterprise, an Owner enables fast mode at Organization settings, then Claude Code. Console organizations are a different path: an administrator enables it in Claude Code preferences on the Anthropic platform, and Anthropic notes that fast mode is a research preview, so the organization also needs fast mode access provisioned. Without that provisioning the API rejects each fast mode request with a 429, which Claude Code treats as a fast mode rate limit, and the rejections continue until access arrives rather than clearing on a cooldown.

You see this sentence and the organization has it enabled

Anthropic documents this as a gateway symptom, and the two configurations are specific. One is authenticating with ANTHROPIC_AUTH_TOKEN alone, with no claude.ai login, no API key and no cached successful check, where Claude Code treats fast mode as disabled without sending the request. The other is a proxy that intercepts the check and answers with its own page, such as a TLS-inspecting proxy returning an HTTP 200 block page, which Claude Code reads as a response saying the organization has fast mode disabled. The setting for this pair is CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK=1, and the network setting does not apply to either of them.

One limit applies to either skip setting. They change only the check on your side. When the organization has fast mode disabled, the API rejects fast mode requests whether or not they are set, and Claude Code then retries at standard speed, turns fast mode off, and reports that the organization disabled it.

Sources: Anthropic, fast mode requirements, enabling fast mode for your organization and fast mode has been disabled by your organization.

"Fast mode is currently unavailable", or a network message

Fast mode is currently unavailable

Fast mode unavailable due to network connectivity issues

What it means. Both come from the availability check rather than from your entitlements, and they mean different things. The second says the check ran and could not reach Anthropic, and requests then run at standard speed. The first says the check was never sent and no earlier success is cached, which is what happens behind an LLM gateway in two named configurations.

When the check could not run

Anthropic names two gateway configurations where the check is never sent: you point ANTHROPIC_BASE_URL at the gateway and set ANTHROPIC_API_KEY to a gateway-issued key that is not in the format of an Anthropic API key, or you set CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC so Claude Code skips the check. Either one produces the "currently unavailable" sentence, and the documented repair for both is CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS=1.

When the network itself is blocked

The check goes to api.anthropic.com and does not follow ANTHROPIC_BASE_URL. It does use a configured HTTP proxy, so the check fails only where api.anthropic.com is unreachable even through the proxy. The documented repair is to allowlist direct egress to api.anthropic.com, or to set CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS=1, which treats a failed check as available while still honouring a "disabled by your organization" answer.

Sources: Anthropic, fast mode behind proxies and LLM gateways.

"Fast mode requires usage credits"

Fast mode requires usage credits

Fast mode unavailable during evaluation. Please purchase credits.

What it means. Usage credits are a billing setting of their own, separate from the plan. Anthropic requires them turned on for a Pro, Max, Team or Enterprise account before fast mode can run, and states plainly that fast mode usage draws directly from those credits even when the plan still has usage left.

Turning them on

• On Pro and Max, turn them on in the Usage credits section of Settings, then Usage, on claude.ai. Running /usage-credits opens that page.
• On Team and Enterprise, a member with billing access turns them on for the organization at Organization settings, then Usage. A member without it runs /usage-credits to send the admins a request.
• On Claude Console, accounts pay per token instead, and do not use usage credits. On the free Evaluation plan the second message above appears, and purchasing credits in Console billing clears it.

What happens when the credits run out mid-session

Anthropic documents the behaviour as a retry rather than a stop. Each rejected fast mode request is tried again at standard speed and pricing, so work continues and there is no cooldown. In an interactive session a notification reads Fast mode disabled · usage credits exhausted and fast mode turns off for the rest of that session, while the saved preference is untouched, so /fast turns it back on later. In a non-interactive run with stream-json output, and through the Agent SDK, the same text arrives as a system notification once per turn and fast mode stays on, which needs Claude Code 2.1.221 or later.

Sources: Anthropic, fast mode requirements and handling rate limits.

What fast mode costs, and where the spend appears

The price. Fast mode is billed per token at a higher rate than standard Opus, and the rate depends on the model. Anthropic publishes 8 dollars per million input tokens and 40 per million output on Opus 5.5, and 10 and 50 on Opus 5 and Opus 4.8. The fast mode price is flat across the full million-token context window.

Why enabling it late costs more

The first time fast mode is enabled in a conversation, Anthropic charges the full fast mode uncached input price for the entire conversation context. The deeper into a conversation that happens, the larger the charge, so enabling it at the start of a session is the cheaper arrangement. Fast mode and standard speed keep separate prompt caches, which means each later switch reads part or all of the conversation uncached again.

Where to find the figure

Run /status first, because the answer depends on how you signed in. A Login method row naming a Claude plan means the usage is billed from subscription usage credits, and the Usage credits section of Settings on claude.ai shows the month's spend including fast mode without separating it out. On Team and Enterprise the organization pays, and /usage shows your own usage-credit spend. A Console account is different again: the Usage and Cost pages on the Anthropic platform separate fast mode when you group by Speed, and that option appears only for a date range that includes fast mode usage.

Its own rate limit

Fast mode has a separate rate limit from standard Opus, and every supported Opus model draws from the same fast mode pool. When the limit is reached, Anthropic documents four steps that happen without input: fast mode falls back to standard speed, the arrow icon beside the prompt turns grey, work continues at standard speed and pricing, and fast mode comes back on its own when the cooldown expires. Running /fast turns it off instead of waiting.

Sources: Anthropic, the cost tradeoff, where fast mode spend appears and handling rate limits. To price a session before you run it, the Claude Code cost calculator takes the model and the token counts.

It follows your model, and it is not an effort level

What it means. Fast mode is bound to the model the session runs on, so a model change can switch it off, and it is a different control from the effort level people often reach for instead. Anthropic treats the two as separate settings that can be combined.

Model switches in both directions

• Switching to a model that does not support fast mode turns it off. Opus 4.7 is the case Anthropic calls out, and before 2.1.221 fast mode stayed on after that switch and the API rejected the requests.
• Switching back to a supported Opus model turns fast mode on again when the saved preference is on. A saved preference that is off is never overridden by a model switch.
• With fastModePerSessionOptIn configured, switching back does not turn it on either, and /fast is the way back.
• Either direction prints Fast mode ON or Fast mode OFF, and the arrow icon shows while fast mode is on. This holds whether the switch came from /model, from /config model=<model> or from a device connected through Remote Control.

Fast mode is not a lower effort level

Anthropic separates them by what they cost you. Fast mode keeps the model's quality and cuts latency while raising the price. A lower effort level cuts the thinking time, which is faster and can be worse on a complex task. They can be used together when speed matters more than depth on straightforward work.

That split also decides when fast mode is the wrong setting. Anthropic suggests it for rapid iteration, live debugging and work with a deadline, and suggests standard speed for long autonomous runs, batch and CI work, and anything where cost matters more than waiting.

Sources: Anthropic, switching models while fast mode is on, fast mode against effort level and when to use it.

What the issue tracker shows

Anthropic's fast mode page is the only page in its documentation index dedicated to the feature. The reports cited here fall into two groups: a billing complaint the documentation answers, and a missing toggle that was a real defect in Anthropic's own catalog.

Fast mode is billed from usage credits, not from the plan's allowance. Anthropic lists subscription plans as a supported route to fast mode, and the condition is usage credits rather than the plan itself. What people object to is narrower and is accurate: fast mode draws from usage credits only and is not included in the subscription rate limits. Two open requests ask for that to change, one asking for fast mode to draw from plan usage or the monthly API credits rather than usage credits only, and one arguing that excluding fast mode from subscription plans leaves Max capacity unable to cover the feature people would spend it on. A third, asking for /fast to consume included quota rather than billing separately, was closed as not planned. All three are requests for a change, not descriptions of documented behaviour.

The missing toggle for Opus 5.5 was a catalog defect. A report that the fast mode toggle was absent for Opus 5.5 in the Claude Desktop app's Code tab traced it to the model catalog rather than to the model. The published catalog gave claude-opus-5 a fast_mode toggle block and claude-opus-5-5 none, so that tab had no switch to draw while /fast in the command line worked on Opus 5.5. The reporter's reading, that this was a catalog gap rather than a model limitation, was right, and the issue was closed as completed on 2026-10-09. That matters when you are staring at a switch that is not there: a surface can lack the toggle for a model that supports fast mode, and no setting of yours explains it.

Sources: the Claude Code issue tracker, read on October 26, 2026: issue 83302 (open), issue 100887 (open), issue 31880 (closed as not planned) and issue 96221 (closed as completed). The page count comes from Anthropic's documentation index.

What does not help

• Reinstalling. Every refusal on this page is a decision about the account, the organization or the model. A fresh copy of Claude Code makes the same request and gets the same answer.
• Retyping /fast. A settings or entitlement refusal does not lift by repetition. It lifts when the setting behind it changes.
• Switching to Sonnet or Haiku to get it back. Fast mode runs on Opus only, and a model switch to a model that cannot support it turns fast mode off rather than restoring it.
• Setting the skip variables and expecting them to bypass an organization. Anthropic states that they change only the check on your side, and that the API still rejects the request when the organization has fast mode disabled.
• Waiting out a Console access refusal. A 429 for an unprovisioned Console organization is not the rate limit cooldown, and Anthropic says the rejections continue until access is provisioned.

Common questions

Why is fast mode not available in Claude Code?

The most common documented reasons are these, and each one says something different. Fast mode runs on Opus only, so the surfaces that show a switch only for a supporting model leave it out on Sonnet, Haiku or another model, while a terminal session switches to Opus when you turn fast mode on. It is not offered on Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or Claude Platform on AWS. A Team or Enterprise organization has it switched off until an Owner enables it. A subscription account needs usage credits turned on. And a network that blocks api.anthropic.com fails the availability check that runs before the toggle appears.

Does fast mode use my Claude plan's included usage?

No. On a Pro, Max, Team or Enterprise plan, Anthropic bills fast mode from usage credits only, and it is not included in the subscription rate limits. Anthropic's fast mode page states that fast mode usage draws directly from usage credits even when the plan still has usage left. Requests that ask for plan usage instead are open on the Claude Code issue tracker, and one earlier request of the same kind was closed as not planned. All are requests for a change, not documented behaviour.

Why does /fast say fast mode requires usage credits?

Usage credits are a billing setting separate from the plan itself. On Pro and Max you turn them on in the Usage credits section of Settings, then Usage, on claude.ai, or by running /usage-credits. On Team and Enterprise a member with billing access turns them on for the organization, and a member without it runs /usage-credits to send the admins a request.

Is fast mode a different model?

No. Anthropic describes fast mode as Claude Opus under a different API configuration that prioritises speed over cost efficiency, with identical quality and capabilities and responses up to 2.5 times faster. It is supported on Opus 5.5, Opus 5 and Opus 4.8, and it is not available on Sonnet, Haiku or other models.

Why did fast mode turn itself off?

The most common documented causes are these. Switching to a model that does not support fast mode turns it off, and that includes Opus 4.7, where fast mode was deprecated on June 25, 2026 and removed on July 24, 2026. The fast mode rate limit is the second: when it is reached, Claude Code falls back to standard speed and pricing, greys the arrow icon beside the prompt, and re-enables fast mode when the cooldown expires. The third is usage credits running out, where fast mode turns off for the rest of an interactive session. The fourth is an organization that has fast mode disabled, where Claude Code retries the request at standard speed and turns fast mode off.

Is ANTHROPIC_SMALL_FAST_MODEL the same as fast mode?

No. ANTHROPIC_SMALL_FAST_MODEL is a deprecated variable for the cheap model behind background work, and Anthropic's model configuration page names ANTHROPIC_DEFAULT_HAIKU_MODEL as its replacement. It has nothing to do with fast mode, which is the /fast toggle on Opus.

Omniscio has no fast mode setting

Omniscio is the desktop app we make for running many Claude Code sessions at once, and it does not have a fast mode switch.

It also does not offer a way to turn fast mode on from inside a session. Omniscio runs Claude Code in non-interactive mode, and Anthropic's rule for that mode is that /fast works only in a session launched with fast mode already in its settings value, and that elsewhere the command reports fast mode is not available. Fast mode therefore applies to sessions outside Omniscio, not to Omniscio's sessions: it turns on in that mode only when the launch settings carry it, and Omniscio does not launch sessions that way.

One setting in the app looks like it could be that switch and is not. Omniscio has a Utility AI model picker for the cheap model behind small background work. Its own description of that work is titles, suggestions, summaries and digests, and it has nothing to do with the speed of a Claude Code session. If you went looking for fast mode in a model list and found that instead, this is the difference.

What the app does help with is the case where a group of sessions is affected at once. Every session carries one of eleven statuses shown in the sidebar, and the ones that need a decision from you collect in an inbox at the top of it. The features page covers the rest of the app, and the download page has the installers.

Sources

Every message, price and setting on this page rests on one of these, read on October 26, 2026. Anthropic marks fast mode as a research preview and says the feature, pricing and availability may change, so read the first item again before you trust a copy of this page from a few months back.

• Anthropic, speed up responses with fast mode, for what fast mode is and is not, the supported models, the toggle and its settings, the defaults by version, the price table, the separate caches, where the spend appears, the effort-level comparison, the requirements list, the four organization blocks, the proxy and gateway cases, the skip settings, the rate limits and the credits behaviour.
• Anthropic, model configuration, for the allowlist interaction with /fast and for ANTHROPIC_SMALL_FAST_MODEL being deprecated in favour of ANTHROPIC_DEFAULT_HAIKU_MODEL.
• Anthropic's documentation index, the list of 264 pages the "only page dedicated to fast mode" line rests on.
• The Claude Code issue tracker for the reports discussed under "What the issue tracker shows": 83302, 100887, 31880 and 96221.