Dev.to AI ๐Ÿค– Ai ๐Ÿ‘ 0 ๐Ÿ“– 12 min read

Claude Code Fast Mode Explained (Opus 4.8 Ends Nov 30)

Claude Code fast mode is a setting that makes Claude write its answers up to 2.5 times faster. You pay a higher price per token for that speed. A token is a small piece of text, roughly a short word or part of a word. T

Claude Code fast mode is a setting that makes Claude write its answers up to 2.5 times faster. You pay a higher price per token for that speed. A token is a small piece of text, roughly a short word or part of a word.

Think of express delivery. It is the same package with the same contents. It arrives faster and costs more. You pick it when waiting hurts, not for every order. This guide covers what it costs, how to switch it on, and what changes for Opus 4.8 users.

Deadline: if you use fast mode on Opus 4.8, it stops working on November 30, 2026. Switch to Opus 5.5 to keep fast mode, or turn fast mode off.

Quick answer

  • Claude Code fast mode makes the same model write up to 2.5x faster, at a higher cost per token.
  • It does not make Claude smarter. The answers are the same quality as standard speed.
  • On subscription plans it uses usage credits only. It is not covered by your plan's normal limits.
  • Opus 4.8 fast mode was deprecated on October 30, 2026. It stops working on November 30, 2026.
  • To keep fast mode, run /model and choose Opus 5.5. To stop using it, run /fast off.

Key words in simple English

  • Model: the AI "brain" that writes the answers. Opus 5.5 and Fable 5.2 are models.
  • Token: a small piece of text. Claude reads and writes in tokens, and prices are set per token.
  • Output speed (tokens per second): how fast Claude writes its answer. Higher means faster text on your screen.
  • Fast mode: a setting that raises output speed and raises the price per token.
  • Usage credits: a pay-as-you-go balance on top of a subscription plan.
  • Rate limit: a cap on how much you can use something in a given time.
  • Prompt cache: text Claude has already read and can reuse more cheaply.
  • API: a way for software to talk to Claude directly.
  • Deprecated: still works, but on its way out.
  • Removed: no longer works.

What is Claude Code fast mode?

Claude Code fast mode is express delivery for your coding assistant. The package (the answer) is the same. It just arrives sooner, and the shipping costs more.

Anthropic's docs call it "the same model with a different API configuration that prioritizes speed over cost efficiency." In plain terms: nothing about Claude changes except how fast the words arrive and how much each one costs.

Here is what gets faster. Fast mode raises output tokens per second, so the answer streams faster once it starts. It does not cut the wait before the first word appears. If the pause before Claude starts typing bothers you, fast mode does not remove it.

Quality does not change. It is the same model, so the answers are not smarter and not worse. For better answers, pick a stronger model. My guide on choosing between Fable 5.2 and Opus 5.5 covers that choice.

Opus 4.8 fast mode ends on November 30, 2026

Anthropic deprecated fast mode for Opus 4.8 on October 30, 2026. It is removed on November 30, 2026. From that day, fast mode requests to Opus 4.8 are turned away.

The Claude Code changelog (v2.1.316, October 31, 2026) says that after removal, fast mode requests on Opus 4.8 return an error. They do not fall back to standard Opus 4.8. The express lane closes, and your package does not quietly go by regular post.

You have two options in Claude Code:

  1. Switch to Opus 5.5. Run /model and pick Opus 5.5. Fast mode stays on. You need Claude Code v2.1.280 or later.
  2. Turn fast mode off. Run /fast off. You keep using Opus 4.8 at standard speed.

Opus 5.5 is also cheaper in fast mode. Opus 4.8 fast costs $10 / $50 per 1M tokens (input / output). Opus 5.5 fast costs $8 / $40. So the move saves money per token and keeps the speed.

One note for developers. Opus 5.5 has its own API differences. If you call the API in code, read my Opus 5.5 migration guide before you switch.

Which models support fast mode (and what it costs)

In simple words: fast mode costs about double the standard price on every model that supports it. Prices are per 1 million tokens, split into input (what you send) and output (what Claude writes).

Model Fast mode price per 1M tokens (input / output) Standard price Status
Claude Fable 5.2 $16 / $80 $8 / $40 Available
Claude Mythos 5.2 (verified organizations only, API) $16 / $80 $8 / $40 Available
Claude Opus 5.5 $8 / $40 $4 / $20 Available
Claude Opus 5 $10 / $50 $5 / $25 Research preview (needs access)
Claude Opus 4.8 $10 / $50 $5 / $25 Deprecated, removed Nov 30, 2026

Fast mode prices apply across the full 1M context window. Prompt caching multipliers apply on top.

Models that do not support fast mode:

  • Sonnet and Haiku models
  • Opus 4.7 (switching to it turns fast mode off)

Claude Code v2.1.280 and later default to Opus 5.5 for fast mode. Fable 5.2 fast mode needs v2.1.321 or later.

How to turn Claude Code fast mode on and off

  1. In the terminal: type /fast. Press Space to toggle it on or off. Then press Enter to confirm. When fast mode is on, you see a "โ†ฏ" icon and the text "Fast mode ON".
  2. In the settings file: set "fastMode": true in your user settings. If your model does not support fast mode, you stay on that model and fast mode stays off.
{
  "fastMode": true
}
  1. In non-interactive mode: this runs Claude Code from a script, with no chat window. It needs Claude Code v2.1.205 or later. Plain /fast in -p mode reports that fast mode is not available, so pass the setting like this:
claude -p --settings '{"fastMode": true}' "Explain this error log in plain English"
  1. In cloud sessions: type /fast on.
  2. To turn it off: run /fast off. To disable it completely, set this environment variable (a setting your computer passes to programs):
export CLAUDE_CODE_DISABLE_FAST_MODE=1

Admins can set fastModePerSessionOptIn: true, which makes every session start with fast mode off.

Who can use it and how it is billed

In simple words: on subscription plans you pay for fast mode with usage credits, not with your plan's normal allowance. On the API you pay per token as usual.

Subscription plans. This covers Pro, Max, Ultra, Team and Enterprise. Fast mode is paid for with usage credits only. It does not count toward your plan's normal limits. It also takes credits straight away, even if you still have plan usage left. If usage credits are off, /fast shows "Fast mode requires usage credits". For how plan limits work, see my weekly limit guide.

Team and Enterprise. An Owner must enable fast mode first. They do it under Organization settings > Claude Code.

Claude Console (API) users. You pay per token. Opus 5 and Opus 4.8 need provisioned access.

The free Evaluation plan. Fast mode is not available. /fast shows "Fast mode unavailable during evaluation. Please purchase credits."

Providers in Claude Code.

  • Google Cloud (Vertex): Fable 5.2 only. It needs the region global, us or eu, and Claude Code v2.1.322 or later. Otherwise you see "Fast mode needs Vertex region global, us or eu".
  • Amazon Bedrock, Microsoft Foundry and Claude Platform on AWS: not available in Claude Code. /fast reports "Fast mode is not available on this provider".

What fast mode really costs

In simple words: fast mode roughly doubles the price per token. The surprise is a one-time bill when you switch it on in a long conversation.

Prices are per 1M tokens, so 500,000 tokens is 0.5.

Example 1. One request with 500,000 input tokens and 50,000 output tokens, no caching.

  • Opus 5.5 standard: 0.5 ร— $4 + 0.05 ร— $20 = $2.00 + $1.00 = $3.00
  • Opus 5.5 fast: 0.5 ร— $8 + 0.05 ร— $40 = $4.00 + $2.00 = $6.00
  • Opus 4.8 fast (until Nov 30): 0.5 ร— $10 + 0.05 ร— $50 = $5.00 + $2.50 = $7.50
  • Fable 5.2 fast: 0.5 ร— $16 + 0.05 ร— $80 = $8.00 + $4.00 = $12.00

Example 2: the hidden cost of turning it on mid-conversation. Say your Claude Code conversation already holds 400,000 tokens of context. Context is everything Claude has read so far in this chat.

  • At standard speed, Claude can usually reread that context from the cache. On Opus 5.5 that costs 0.4 ร— $0.20 = $0.08.
  • The first fast mode request pays the full uncached fast price for the whole context: 0.4 ร— $8 = $3.20. That is before Claude writes a word.

That is 40 times more for the same reread. Anthropic's docs confirm it: the first time you turn fast mode on in a conversation, the whole context is billed at the full fast input price. Fast and standard speed keep separate caches, so switching back and forth repeats this cost.

My rule: turn fast mode on at the start of a session, not in the middle of a long one.

When to use fast mode (and when not to)

Use fast mode when waiting hurts:

  • You are working live and every pause breaks your focus.
  • You are debugging with Claude in a tight back-and-forth loop.

Skip it when waiting is fine:

  • The task is long and you can leave it running.
  • You are deep into a long conversation (see the hidden cost).
  • A cheaper model could do the job. My Opus vs Sonnet vs Haiku guide shows when a smaller model is enough.

Many slowdowns have free fixes. Try my slow Claude Code fix guide before you pay for the express lane.

Fast mode messages and how to fix them

In simple words: most messages mean something is switched off, missing, or busy.

Message you see What it means What to do
Fast mode requires usage credits Your plan has no usage credits turned on. Turn on usage credits.
Fast mode has been disabled by your organization. An admin turned it off. Claude Code retries at standard speed. Ask your Owner or admin.
Fast mode is not available on this provider You use Bedrock, Microsoft Foundry or Claude Platform on AWS. Use the Anthropic API for fast mode.
Fast mode unavailable during evaluation. Please purchase credits. You are on the free Evaluation plan. Buy credits.
Fast mode needs Vertex region global, us or eu Your Google Cloud region is not supported. Use global, us or eu, and v2.1.322 or later.
is not in your organization's allowed models /fast would switch to an Opus model your organization blocks. Ask an admin, or pick an allowed model.
Fast mode unavailable due to network connectivity issues A network, gateway or proxy problem. Check your connection. For gateways, the env vars CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS=1 and CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK=1 exist.
Fast mode is currently unavailable Fast mode is temporarily not available. Try again later.
Fast mode paused: your cloud provider refused a fast request ยท trying again in 30m Your cloud provider refused a fast request. Wait. Claude Code tries again on its own.
Using standard speed on Fable 5.2 ยท fast mode wasn't available You got standard speed instead. Nothing needed.
Fast mode disabled ยท Out of usage credits You ran out of credits. There is no cooldown. Add usage credits.

When you hit fast mode's own rate limit, Claude Code falls back to standard speed. The "โ†ฏ" icon turns gray during a cooldown. Fast mode then comes back on automatically.

After November 30, 2026, expect Opus 4.8 fast requests to fail with an error. They will not fall back to standard speed. The exact error text has not been published yet, so I cannot quote it. I will add it here once it is public.

If you are not a developer, the sections above give you the full picture. The next section is for people calling the Claude API directly.

Fast mode in the Claude API

The speed parameter. Add speed: "fast" to a Messages API request. The API docs promise up to 2.5x more output tokens per second than standard speed.

Beta header rules. A beta header is a flag that opts you in to an early feature. The header fast-mode-2026-02-01 is needed only for Claude Opus 5 and Claude Opus 4.8, through the beta messages endpoint. Opus 5.5, Fable 5.2 and Mythos 5.2 need no beta header.

Check what you got. The response field usage.speed reports "fast" or "standard".

Before (Opus 4.8 fast mode, which stops working on November 30, 2026):

import anthropic

client = anthropic.Anthropic()

response = client.beta.messages.create(
    model="claude-opus-4-8",
    max_tokens=4096,
    speed="fast",
    betas=["fast-mode-2026-02-01"],
    messages=[{"role": "user", "content": "Summarize this stack trace in three bullet points."}],
)

From November 30, 2026, any request to claude-opus-4-8 with speed: "fast" returns an error.

After (Opus 5.5 fast mode, no beta header):

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=4096,
    speed="fast",
    messages=[{"role": "user", "content": "Summarize this stack trace in three bullet points."}],
)

print(response.usage.speed)  # "fast" or "standard"

Three things changed: the model name, the call (client.messages, not client.beta.messages), and the betas line is gone. The last line prints which speed you got. For other changes, see the migration guide.

Rate limits and 429 errors. Fast mode has its own rate limit. A 429 error means "too many requests". It includes a retry-after header that tells you how long to wait. Response headers also show what is left: anthropic-fast-input-tokens-remaining and anthropic-fast-output-tokens-remaining. The SDK retries fast mode 429s twice by default. To fall back to standard speed, retry the request without speed.

The cache note. Fast and standard requests do not share cached text. When you fall back from fast to standard, that request misses the cache and loses the cache discount. A fallback is not free.

Not available with the Batch API, or a Priority Tier commitment.

Managed Agents and Foundry. In Managed Agents, set speed on the agent's model object. In Microsoft Foundry, use separate model IDs that end in -fast and leave out speed.

Frequently asked questions

What is Claude Code fast mode?

It makes the same Claude model write up to 2.5x faster, at a higher price per token. Turn it on with /fast.

Does fast mode make Claude smarter?

No. It is the same model, so answers are not better or worse. It also does not cut the wait before the first word.

Is fast mode included in my Pro or Max plan?

No. It is available through usage credits only, and it draws from credits even if you have plan usage left.

How much does fast mode cost?

On Opus 5.5 it costs $8 input and $40 output per 1M tokens, against $4 and $20 at standard speed. Turning it on mid-conversation also adds a one-time full-price read of your context.

What happens to Opus 4.8 fast mode on November 30, 2026?

It is removed. Fast requests return an error and do not fall back to standard speed. Switch to Opus 5.5 with /model, or run /fast off.

Which models support fast mode?

Fable 5.2, Opus 5.5, Opus 5 and Opus 4.8 (until November 30, 2026) on the Anthropic API. Mythos 5.2 works for verified organizations on the API. Sonnet, Haiku and Opus 4.7 do not.

Why does Claude Code say "Fast mode requires usage credits"?

Usage credits are not turned on for your subscription. Turn them on in your account. On Team and Enterprise, an Owner must also enable fast mode.

Can I use fast mode on Amazon Bedrock?

Not in Claude Code. /fast reports "Fast mode is not available on this provider".

Is fast mode worth it?

It depends on whether you are watching. If you sit and wait for each reply, for example while debugging live, paying about double for up to 2.5x the speed can be worth it. If the task runs while you do something else, or a cheaper model could handle it, standard speed gives you the same answer for half the price.

Where to go next

Sources: Claude Code docs: Fast mode, Claude API docs: Fast mode, Claude Code changelog

Last updated: November 20, 2026. I will update this page after the Opus 4.8 removal on November 30, 2026.

๐Ÿ“ฐ Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes โ€” full credit and traffic to the original publisher.