What is fast mode? Claude Code, Codex, Cursor, and Grok

Fast mode is a speed setting on a model the product already offers. You pay more for the quicker replies. It is not a separate model you install.

The price is not one number across tools. Claude Code publishes a per-token price and says the quality stays the same. Codex multiplies plan usage. Cursor prices Fast per model, and the multiple is not always 2x. Grok Build’s command list has no fast toggle.

How fast mode works

You turn the setting on for a model that advertises it. The next tokens come back with higher priority or a faster configuration. The bill goes up, either as a higher token price or as more of your plan’s usage for the same work.

Turning it on in the middle of a long chat can cost more than turning it on at the start. Claude Code bills the whole conversation as uncached input the first time you enable it, because fast mode and standard speed keep separate caches. A later switch pays that uncached read again.

Reasoning effort is a different knob. Lower effort means the model thinks less, and the answer can get worse. Fast mode, on Claude Code’s page, keeps the same quality and charges more. You can use both.

Fast mode in each tool

Read the column for the tool you have open. A 2.5x figure in one column is not the speed of another.

Claude Code

Fast mode is a configuration for Claude Opus, not a different model. The docs say you get identical quality and capabilities, with responses up to 2.5x faster, at a higher cost per token. It runs on Opus 5.5, Opus 5, and Opus 4.8. It does not run on Sonnet, Haiku, or other models. Fast mode for Opus 5.5 is available. Fast mode for Opus 5 and Opus 4.8 is a research preview.

Toggle it with /fast, then Space, then Enter. Or set "fastMode": true in your user settings. An interactive toggle persists across sessions unless you set fastModePerSessionOptIn to true, which starts every session off.

You can run /fast while Claude is working. The running turn finishes at the old speed. The change applies on the next turn. If the current model does not support fast mode, turning it on switches you to Opus. Claude Code shows “Fast mode ON” and a ↯ icon. Turning it off leaves you on Opus. /model is how you leave that model.

Prices per million tokens, input then output, are $8 and $40 for Opus 5.5, and $10 and $50 for Opus 5 and Opus 4.8. The price is flat across the full 1M context. On Pro, Max, Ultra, Team, and Enterprise, fast mode is billed from usage credits. It is not included in the subscription rate limits. Team and Enterprise organizations have it off until an Owner enables it. It is not available on Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry, or Claude Platform on AWS.

Fast mode for Opus 4.8 was deprecated on October 30, 2026, and will be removed on November 30, 2026. If you are still on Opus 4.8 with fast mode on after that date, the API rejects the requests. Keep fast mode by switching to Opus 5.5 with /model, or stay on Opus 4.8 at standard speed with /fast off. Opus 4.7 does not support fast mode, so switching to it turns the setting off.

When you hit the fast-mode rate limit, requests fall back to standard speed, the ↯ icon turns gray, and fast mode comes back when the cooldown ends. If usage credits run out during an interactive session, Claude Code turns fast mode off for the rest of that session. The saved preference does not change.

See Claude Code fast mode.

OpenAI Codex

Fast mode speeds up GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Astra, GPT-6 Sol, and GPT-6 Luna, where your account has them. For GPT-5.6 the speed increase is 1.5x. The speed page does not state a speed multiple for the GPT-6 models.

/fast toggles the current model’s Fast service tier and saves the selection. If the model does not advertise a Fast tier, Codex does not show /fast. /statusline can show the setting in the footer. That row is the status line.

To keep it on for new turns, set both of these in config.toml:

service_tier = "fast"

[features]
fast_mode = true

service_tier is the preferred tier for new turns. The value fast maps to the request value priority. features.fast_mode turns the Fast controls on. The config reference says that flag is on by default, and that availability still depends on the model and the account. It is separate from features.ultrafast_mode.

On supported models, included subscription limits are charged at 2.5x the Standard rate. Purchased credits and Enterprise pay-as-you-go are billed at 2x Standard. Fast mode is available in the ChatGPT desktop app, the Codex CLI, and the IDE extension when you sign in with ChatGPT. With an API key, Codex uses API token pricing, and those ChatGPT credit multipliers do not apply.

Ultrafast is a different tier. It supports GPT-6 Astra and GPT-6.1 Sol. GPT-6 Astra Ultrafast generates tokens up to 8x faster than GPT-6 Astra Standard in Codex. Included limits for Ultrafast are 8x Standard. Credits and Enterprise pay-as-you-go are 6x. It is available in Codex and ChatGPT Work on Pro $500 and on eligible Enterprise and Edu plans. Enterprise has it off until a workspace owner enables it. Do not read the /fast 2.5x line as the Ultrafast price.

See Codex speed and Codex slash commands.

Cursor

Fast is a priced tier on specific models. The pricing page does not give one multiplier for every model.

Grok 4.7 additional usage, per million tokens, is $2 input, $0.50 cached input, and $6 output at standard speed. Fast is $4, $1, and $12. Past 256k input tokens, standard requests bill at 2x those rates and Fast requests bill at 3x the standard rates, up to 500k. On Premium and higher, and on legacy Pro, Pro+, and Ultra, Fast is the default speed for Grok 4.7. On Plus, and on the legacy Start plan in India, Grok 4.7 stays at medium effort and standard speed.

Composer 2.5 is $0.50 input, $0.20 cached, and $2.50 output. Composer 2.5 Fast is $3, $0.50, and $15. That is not a 2x tier.

Opus 5.5 bills at $4 input and $20 output per million. The Fast tier id is claude-opus-5-5-fast. The model page calls it higher-priority output, at $8 and $40. On legacy request-based plans it requires Max Mode. Cursor does not say the quality is identical. That sentence is on Claude Code’s page only.

The Plus plan runs Cursor’s own models with Fast off. The pricing FAQ says you upgrade to Premium or higher to choose Fast mode and effort levels.

The CLI changelog says /fast toggles Fast when the current model supports it, and that the shortcut can be disabled under Model slash commands in /config. Choosing /fast no longer leaves Max Mode stuck on after you pick a model that does not need it. An explicit /max-mode choice stays on. The CLI slash-command reference does not list /fast. The changelog is the page that does.

Prices above were checked on the Cursor pricing and model pages in November 2026.

See Cursor models and pricing, Grok 4.7, Claude Opus 5.5, and the CLI changelog.

Grok Build

The modes and commands page lists no /fast command and no fast-mode setting. There is nothing to configure from that page.

See Grok Build modes and commands.

Details checked against each vendor’s documentation in November 2026.

A real example

Two configs that the vendor pages show. Claude Code, in user settings:

{
  "fastMode": true
}

Codex, in config.toml, for new turns on a model that advertises Fast:

service_tier = "fast"

[features]
fast_mode = true

In either CLI, /fast is the toggle for the session you are in. In Cursor, /fast is the CLI shortcut from the changelog, and the Plus plan does not offer the choice.

When to use fast mode

Use it when you are waiting on the next reply. Live debugging, a tight edit loop, and a deadline are the cases Claude Code names. Codex is the same trade: quicker answers, and included usage counted at 2.5x.

Leave it off for a long unattended run, a batch job, or anything where the bill matters more than the wait. Claude Code says standard speed is the better fit there. On Cursor’s Plus plan you do not get the choice. On a model with no Fast tier, Codex hides /fast and Claude Code will switch you to Opus if you turn it on.

How tokens are counted is the bill this setting multiplies. Prompt caching is why Claude’s mid-chat switch costs a fresh read.

Pitfalls

The 2.5x figures are not the same unit. Claude Code says responses can be up to 2.5x faster. Codex says included limits are charged at 2.5x Standard, and it states a 1.5x speed increase only for GPT-5.6. Cursor’s Grok 4.7 Fast tier is 2x the token price, and Composer 2.5 Fast is several times that. Copying one number onto another tool will mis-bill the run.

Claude Code and standard speed do not share a cache. The first time you enable fast mode, you pay fast-mode uncached input for the whole conversation. Each later switch pays again. Enable it at the start if you already know you want it.

Opus 4.8 fast mode ends on November 30, 2026. After that, requests fail if that model is still selected with fast mode on. Ultrafast in Codex is not /fast. Its included-limit multiple is 8x, and its credit multiple is 6x.

A gray ↯ in Claude Code means the fast-mode rate limit is cooling down and you are on standard speed for now. Running out of usage credits turns fast mode off for the rest of that interactive session, and your saved preference stays on.

Related on flaviocopes.com

• The complete guide to the Codex CLI shows /fast and the 2.5x included-limit rate.
• GPT-6.1 Sol shows the service_tier and fast_mode keys in a Codex config.
• Prompt caching is the cache Claude Code splits between fast mode and standard speed.
• Reasoning effort is the other speed knob, the one that can change the answer.

Related glossary entries

• Reasoning effort: Reasoning effort sets how long a model thinks before it answers. The levels in Claude Code and Codex, what Cursor's plans fix, and how Grok sets effort.
• Tokens and usage limits: A token is the unit a model reads and writes. A usage limit is how much of that work a plan includes. How Claude Code, Codex, Cursor and Grok count it.
• Status line: A status line is the row under a coding session that shows the model, context, or a script you wrote. How Claude Code, Codex, Cursor, and Grok Build set it.

All AI glossary entries →

Want me to talk about your product? You can sponsor this site.