Toggling Fast Mode with /fast in the CLI

/fast is the CLI's switch for Fast mode: type it once to turn the current model's Fast tier on, type it again to turn it off. Codex saves the choice, so it carries into later sessions until you change it. That persistence is the part worth understanding before you use it, because Fast isn't free — and how much it costs depends on how you sign in.

What the command does

/fast toggles a Fast service tier for the active model and persists the selection. It only exists when the model advertises a Fast tier: OpenAI describes the Fast commands as catalog-driven, so if the current model's catalog entry has no Fast tier, Codex doesn't show /fast at all. If the command is missing from the slash popup, check the model before assuming something is broken. Switching models with /model can change whether /fast is offered.

To see the current state without toggling it, add a Fast mode item to the footer with /statusline. It saves you from toggling twice to check, and makes it obvious when a session is running on the premium tier.

Making it a default in config

You can set the preference in config.toml instead: service_tier = "fast" sets the preferred tier for new turns. The related [features].fast_mode flag enables tier selection in the TUI and is on by default. One detail connects to the API's history: in Codex config, fast maps to the request value priority — the tier's name before it was renamed on July 30, 2026 — so request logs from Codex may show the old name.

In managed environments, an administrator can pin Fast mode on or off for local Codex clients through requirements.toml. A user setting can't override that, which explains a /fast that seems to do nothing on a company machine.

The same file serves more than the CLI. The ChatGPT desktop app and the IDE extension read the same config.toml, and Fast mode is available in all three when you sign in with ChatGPT, so a service_tier set there follows you from terminal to editor. That's convenient for a preference and easy to forget for a cost: a Fast default chosen for interactive CLI work also applies to long tasks you start from the desktop app.

What it costs, by how you sign in

Signed in with ChatGPT, Fast uses your plan's included usage at 2.5x the Standard rate, and purchased credits — or Enterprise pay-as-you-go usage — at 2x. OpenAI notes that these are billing multipliers and don't describe the speed increase. So an hour of Fast work drains a plan's allowance faster than the same hour drains credits, and both faster than Standard. Why Fast mode burns your Codex plan faster works through the consequences.

With an API key, Codex bills at API token prices, and ChatGPT's multipliers don't apply. Fast is a separate price row per model on the pricing page, a premium over Standard; the service tiers reference lists them.

What it buys

For GPT-5.6 and GPT-5.5, OpenAI states the speed increase inside Codex as 1.5x. For the GPT-6 models — GPT-6.1 Sol, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna — it says Fast speeds them up where available, without a figure. Either way, Fast speeds up the model, not the rest of the session: time spent running tests, reading files and waiting on tools doesn't shrink. A session that's mostly waiting on a slow test suite gains little from Fast and still pays the multiplier.

When to turn it on

Fast earns its cost when you're watching the output and waiting on it: interactive debugging, a tight edit-and-check loop, a review you want back while you still have the context in your head. It's poor value for anything you'll read later — long unattended runs, overnight refactors, scheduled tasks — where nobody benefits from the speed.

That suggests a simple habit. Leave Standard as your saved default, turn Fast on with /fast at the start of an interactive stretch, and turn it off before handing the session a long autonomous task. Because the setting persists, forgetting the second step is how Fast ends up billing a week of background work.

Measuring whether it helped

The honest test is cheap. Pick a task you run often, run it once on Standard and once with Fast on, and compare two things: how long you actually waited, and how much usage each run consumed. If the Fast run saved a minute on a task where you'd have been doing something else anyway, the multiplier bought nothing. If it turned a stop-and-wait loop into a continuous one, it may be worth every unit. The answer differs by task, which is the argument for toggling per session rather than setting Fast as a permanent default.

Fast and rate limits on an API key

With an API key, Fast shares the model's Standard rate limit rather than adding a separate pool, so it makes each request faster without raising how many you can send. And a sharp ramp in traffic can get some Fast requests downgraded to standard speed and billed at standard rates — mostly a concern for scripted, high-volume use rather than one interactive session, but worth knowing if you drive Codex from automation with Fast set in config.

Fast isn't the top speed any more

GPT-6 Astra also has Ultrafast, a faster and more expensive tier, available in Codex only on the top Pro tier and on eligible Enterprise and Edu plans. Choosing Standard, Fast or Ultrafast compares all three and covers when a cheaper model at Fast speed beats a pricier model at any speed.

Checking what you're spending

/usage shows your account's token activity, and /status shows the session's model and token usage — see checking Codex usage with /status and /usage. If a plan's allowance is disappearing faster than expected, a forgotten /fast is one of the first things to rule out.