Codex Fast Mode: How to Turn Off or On, Toggle Commands & Quota Cost
To turn off Fast mode in the Codex CLI, type /fast and press Enter. The command acts as a toggle: run it once to switch to Fast mode, and run it again to switch back to Standard speed. Codex automatically saves your selection for future turns.
Toggles Fast tier on or off. Selection is saved automatically.
Inspects session tier. Run /statusline to pin it to your footer.
Model & speed settings
Adjust speed controls in desktop Codex model settings. (Desktop-specific view; distinct from Codex Web & CLI).
Fast Mode vs. Standard Mode: Speed & Credit Multipliers
OpenAI documents specific speedups and credit multiples depending on the active model and billing surface:
| Model | Expected Speedup | ChatGPT Credit Cost | API Platform Billing | Surface Availability |
|---|---|---|---|---|
| GPT-6 Sol Current GPT-6 model; availability varies | Faster responses; exact speed multiplier not published | 2.5x Standard (where available) | 2x applicable Standard API rates; longer-input and regional modifiers still apply | Rolling out; check account and client |
| GPT-6 Luna Current GPT-6 model; availability varies | Faster responses; exact speed multiplier not published | 2.5x Standard (where available) | 2x applicable Standard API rates; longer-input and regional modifiers still apply | Rolling out; check account and client |
| GPT-5.6 Sol General coding and reasoning | 1.5x faster (Codex) / up to 2.5x (API) | 2.5x Standard | 2x standard token pricing | Supported |
| GPT-5.5 Previous-generation coding model | 1.5x faster | 2.5x Standard | 2.5x standard token pricing | Supported |
| GPT-5.4 Historical Codex model; retired from ChatGPT sign-in August 31 | 1.5x faster | 2.0x Standard | 2x standard token pricing | Retired from Codex with ChatGPT sign-in; API unaffected |
| GPT-6 Astra Frontier reasoning and complex debugging | Faster than Standard (no exact multiplier published) | 2.5x Standard (where available) | 2x standard ($20/M in, $100/M out) | Supported |
| GPT-5.3-Codex-Spark Research preview on Cerebras WSE-3 hardware (deprecated) | Exceeded 1,000 tok/s (dedicated hardware) | Separate preview rate limits (did not draw from standard limits) | Not in standard API | Separate Model (Not Fast Mode) |
Standard mode uses 1.0x credit metering. Fast mode delivers roughly 1.5x faster generation in Codex (up to 2.5x in the API for GPT-5.6 Sol) with documented 2.0x to 2.5x credit multipliers where credit rates apply. For users relying on included plan allowances, Fast mode accelerates 5-hour and weekly quota depletion per active hour, though OpenAI does not define a universal fixed multiplier for included allowance meters.
What is Codex Fast Mode?
Fast mode is an optional speed tier in Codex that routes requests through higher-throughput, lower-latency serving infrastructure. It accelerates token generation for supported models in exchange for higher credit consumption or per-token API charges.
When signed in with a ChatGPT account, OpenAI documents Fast mode as roughly 1.5x faster for GPT-5.6 Sol, GPT-5.5, and GPT-5.4. For GPT-6 Astra, OpenAI documents Fast mode as faster than Standard without publishing an exact multiplier.
Fast mode is catalog-driven: it is only available when the active model advertises a Fast tier in its catalog definition. If you select a model without a Fast tier, Codex hides Fast mode controls from the command palette.
Codex CLI Controls: Toggle, Verify Status & Customize Footer
In the Codex terminal interface (TUI), control Fast mode directly using built-in developer commands:
Type /fast to turn the Fast tier on; type /fast again to turn it off. Codex saves the selection automatically.
Type /status to view active session details, including current model, approval policy, context tokens, and active service tier.
Type /statusline to configure the TUI footer to permanently display a Fast mode indicator while working.
OpenAI documentation currently presents two command formats: the OpenAI Speed guide describes /fast on, /fast off, and /fast status, whereas the OpenAI Developer Commands reference and current Codex CLI releases expose /fast as a parameterless toggle that switches between Fast and default while saving your selection. In modern builds, run /status to inspect session settings and /statusline to pin the tier indicator to your footer.
ChatGPT Desktop App & Workspace Admin Controls
Codex operates across separate surfaces—including the Codex CLI, the desktop view in the ChatGPT desktop app, and the standalone Codex Web platform. Fast Mode controls and options differ across these surfaces:
In the desktop application, manage speed settings through graphical model controls:
- Speed Controls: Within the Codex view, adjust model speed and toggle Fast mode off or on through the model settings.
- Surface Boundaries: These graphical controls apply to the dedicated Codex view inside the ChatGPT desktop app. OpenAI also operates Codex Web as a separate browser surface alongside the Codex CLI, with control settings varying by environment.
- Turn-by-turn Control: You can adjust speed settings during an active conversation without resetting your session history.
Organization admins configure starting behaviors for managed team members:
- Admin Console: Navigate to Workspace settings > Models to configure starting speed preferences and Fast Mode availability across the organization.
- Role Overrides: Admins can selectively enable or disable Fast Mode for specific team roles.
- Policy Enforcement (
requirements.toml): Enterprise policies can pinfeatures.fast_mode = falseto prevent developers on managed machines from running Fast tier requests.
Persistent CLI Configuration: config.toml Settings
When you run /fast in the CLI, Codex toggles the tier and saves the selection automatically. If you want to explicitly enforce or remove Fast Mode defaults across all new sessions, configure your user configuration file:
# Preferred service tier for new turns ("fast" or "default")
service_tier = "default"
# Optional: default service tier for newly created threads
# models.new_thread.service_tier = "default"The relevant configuration keys in ~/.codex/config.toml include:
service_tier: Preferred service tier for new turns. Set to"fast"to default to Fast mode, or remove the line / set to"default"to ensure Standard speed.models.new_thread.service_tier: Default service tier for new threads. Overridden if an explicit session-level flag is passed.[features] fast_mode: Enables model-catalog service tier selection in the TUI (stable and enabled by default in modern Codex). Setting this tofalsedisables the feature entirely.
Why Is Fast Mode Already Enabled Without Me Turning It On?
If you noticed your quota draining quickly and discovered Fast mode was already active, it is usually due to one of four reasons:
In ChatGPT Business, Enterprise, and Edu workspaces, Fast Mode is often configured as the organization-wide starting speed under Workspace settings > Models. New chats inherit this default unless you explicitly toggle it off.
In the Codex CLI, typing /fast saves your preference. If you toggled Fast mode on during a previous task or debugging session, Codex remembered the preference for subsequent turns.
If your development machine uses shared dotfiles or team templates, service_tier = "fast" may have been populated in ~/.codex/config.toml during initial environment setup.
Selecting a Fast tier or high-speed preset in the desktop application's model controls persists for subsequent turns until changed back to standard speed.
Does Codex Fast Mode Lower Code Quality or Reasoning?
No. Fast mode changes serving speed, not the model's intelligence, architecture, or output quality.
A common concern is whether Fast mode substitutes a smaller model or truncates reasoning steps. OpenAI documents GPT-5.4 Fast mode as the same model and same intelligence, just faster, and explicitly states that GPT-5.6 Sol Fast mode operates with no change in intelligence.
If you set High reasoning effort on GPT-5.6 Sol or GPT-6 Astra, the model still completes its full internal reasoning chain before streaming code changes. Fast mode alters how quickly tokens stream across the network, not how deeply the model evaluates the problem.
Does Codex Fast Mode Use More Quota? (5-Hour & Weekly Impact)
Yes. Codex Fast mode accelerates quota depletion across both your rolling 5-hour limit and your 7-day weekly allowance.
Because tokens stream faster, you complete turns quicker and submit more prompts per hour. Fast tier activity draws down the rolling 5-hour window more rapidly, meaning consecutive intensive turns can exhaust your burst headroom much earlier than in Standard mode. Review sliding mechanics in our Codex 5-hour limit guide.
Where credit-billed usage applies, Fast mode carries a documented 2.0x or 2.5x credit multiplier over Standard. For included weekly allowances, OpenAI notes that quota consumption varies based on task complexity, input/output tokens, reasoning effort, and model tier—with Fast tier requests contributing more heavily toward weekly limits. Learn how weekly reanchoring works in our Codex weekly limit guide.
| Tier | Correctness | Runtime | Observed 5-Hour Movement | Client-Reported Input Tokens |
|---|---|---|---|---|
| Standard | 112 / 112 checks | 92.0s | 2 percentage points | 171,739 |
| Fast | 112 / 112 checks | 82.6s | 8 percentage points | 155,609 |
On ChatGPT Plus, the same frozen coding task at Medium reasoning passed all 112 tests on both tiers. Fast finished about 10.2% faster in wall-clock time and moved the displayed 5-hour meter 8 points against Standard's 2 points. The total wall-clock speedup (10.2%) was smaller than raw model token velocity because fixed client overheads—initialization, tool invocation, test suite execution, and local file diffing—take the same amount of time regardless of serving tier.
This is Codex Usage's own observation from controlled research, separate from official OpenAI documentation. It represents one single run per tier on a whole-point integer meter, with the Standard baseline captured earlier in the same window. The 8:2 meter movement is not a universal Fast-mode multiplier and does not establish a universal rule for included allowance meters. See the canonical Standard vs. Fast measurement record for methodology details.
If your quota is draining unexpectedly fast, check whether Fast mode was left on using our troubleshooting guide for rapid quota drain.
ChatGPT Subscription vs. OpenAI Platform API: Two Distinct Systems
How Fast mode operates and bills depends on your authentication method:
When you sign in with codex login using a ChatGPT Plus, Pro, or Business account, Fast mode draws from your included allowance or purchased credits. Where credit rates apply, turns consume credits at 2.0x Standard for GPT-5.4 and 2.5x Standard for GPT-5.5, GPT-5.6 Sol and GPT-6 Astra, Sol and Luna (where available). On included subscription allowances, Fast tier requests accelerate allowance depletion. Pro $100, $200, and $500 currently have no five-hour usage limit; other plans may have five-hour and weekly meters.
If you authenticate with an OPENAI_API_KEY, requests route to Platform API pay-as-you-go billing. The tier accepts either service_tier: "fast" or "priority", billing per million tokens at model Fast rates (such as 2x Standard on GPT-5.4, GPT-5.6 Sol, and GPT-6 Astra, and 2.5x on GPT-5.5). API usage is completely independent of ChatGPT 5-hour rolling caps.
Compare subscription economics and token rates in our Codex pricing comparison guide and credits vs. resets guide.
GPT-6 Astra Fast Mode
GPT-6 Astra supports Fast mode on eligible ChatGPT plans and the OpenAI Platform API. OpenAI documents it as faster than Standard processing, but does not publish a standalone speed multiplier for Astra.
In exchange for faster generation, Astra Fast mode consumes credits at 2.5x Standard where available for ChatGPT subscribers.
In the developer API, Astra Fast mode is billed at $20/M input tokens and $100/M output tokens—exactly 2x the standard $10/M input and $50/M output rates. For full details on allowances and capacity tiers, see our GPT-6 Astra usage limits guide.
Fast Mode vs. Codex-Spark: Distinct Features
Fast mode and Codex-Spark are different concepts:
- Works on supported general models, including GPT-6 Astra, Sol and Luna where available. GPT-5.4 has retired from Codex with ChatGPT sign-in.
- Documented as roughly 1.5x faster in Codex (up to 2.5x for Sol in the API).
- Consumes ChatGPT credits at 2x or 2.5x Standard where credit rates apply.
- Toggled via
/fastor desktop model controls.
- Separate research-preview model served on Cerebras Wafer Scale Engine 3 hardware.
- Built for extreme speed, generating over 1,000 tokens per second.
- Ran on its own rate limits that did not count against standard limits during its preview.
- Deprecated on September 14, 2026, and removed from Codex client surfaces (CLI, IDE, and desktop app).
For details on Spark's separate limits and hardware setup, see our GPT-5.3-Codex-Spark limits guide.
When Is Codex Fast Mode Actually Worth Using?
Because Fast mode applies higher credit multipliers and accelerates allowance drain, toggling it selectively keeps you from exhausting limits when speed is unnecessary:
- Interactive Pair Programming: When you are actively typing and waiting for real-time code modifications to stay in flow.
- Production Incident Triage: During live outages or urgent bug investigations where saved seconds directly matter.
- Live Demonstrations: When presenting live code generation to team members or stakeholders without lengthy pauses.
- Abundant Flexible Credits: When your account has purchased credits or operates on a dedicated API budget.
- Background & Autonomous Loops: Large codebase migrations, broad test generation, or multi-step subagent execution.
- Large File Summaries: Prompts reading thousands of lines of context where input processing dominates and output speed is secondary.
- Fixed Allowance Budgets: Coding on a standard Plus or Business plan without flexible add-on credits.
- Late in Your 5-Hour or Weekly Cycle: When accelerated allowance consumption risks locking you out before your next reset window. Check your current Codex usage before leaving Fast on.
Frequently Asked Questions About Codex Fast Mode
How do I turn off Fast Mode in the Codex CLI?
In any active Codex CLI session, type /fast and press Enter. The /fast slash command acts as a toggle: running it turns Fast mode off if it was on, or on if it was off. Codex automatically saves the selection across subsequent turns. You can confirm your active tier by running /status.
How do I turn off Fast Mode in the ChatGPT desktop app?
In the ChatGPT desktop app, you can adjust speed or turn off Fast mode in your model settings within the Codex view. (The Codex view inside the ChatGPT desktop app is desktop-specific; OpenAI separately operates Codex Web as a browser surface, and Fast Mode controls can differ by surface).
How do I check if Fast Mode is currently active?
In the Codex CLI, run /status to view your session summary, which displays the active model, approval policy, context size, and service tier. You can also run /statusline to add a persistent Fast-mode indicator directly to your TUI footer.
Why was Fast Mode already enabled on my account?
Fast Mode may be active for three reasons: (1) your workspace admin configured Fast Mode as the starting default under Workspace settings > Models; (2) you previously typed /fast in the CLI, which saved the selection; or (3) your ~/.codex/config.toml contains service_tier = "fast" or models.new_thread.service_tier = "fast".
Does Codex Fast Mode use more credits or quota?
Yes, but credit multipliers and plan allowances operate differently. Where credit billing applies, OpenAI documents a 2.0x credit multiplier for GPT-5.4 and 2.5x for GPT-5.5, GPT-5.6 Sol, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. For included plan allowances (such as rolling 5-hour and weekly meters), Fast mode accelerates quota depletion because turns generate faster and Fast tier requests contribute more to usage, but OpenAI does not publish a universal 2x or 2.5x multiplier for included allowance meters. Actual allowance consumption varies by task complexity, input/output volume, and reasoning settings.
Does Codex Fast Mode produce lower-quality code or reduce reasoning?
No. Fast mode changes serving velocity, not model weights, intelligence, or reasoning depth. OpenAI documents GPT-5.4 Fast mode as the same model and same intelligence, and confirms no intelligence change for GPT-5.6 Sol. If High reasoning effort is configured, the model still completes its full reasoning chain before generating responses.
Does the /fast toggle persist across CLI sessions?
Yes. In modern Codex CLI releases, typing /fast toggles the tier and Codex saves the selection for future turns and sessions. If you want to enforce Standard mode permanently across all projects, check ~/.codex/config.toml and ensure service_tier = "fast" is removed.
How does Codex Fast Mode differ from OpenAI Platform API Fast Mode?
Codex CLI and desktop interfaces operate under ChatGPT subscription plans, where credit-billed usage applies documented 2x to 2.5x multipliers and included allowance meters deplete faster under Fast tier activity. OpenAI Platform API Fast mode is billed separately on a pay-as-you-go per-token basis using the request parameter service_tier: "fast" (or legacy "priority"), completely independent of ChatGPT subscription limits.
The facts, commands, and multipliers on this page are verified against official OpenAI documentation:
- OpenAI Help Article 11369540: Using Codex with your ChatGPT Plan
- OpenAI Help Article 20001516: Managing usage with GPT-6 Astra in Work and Codex
- OpenAI Developer Commands: Toggle Fast Mode with /fast
- OpenAI Documentation: Speed & Fast Mode Guide
- OpenAI Codex Official Pricing & Service Tier Documentation
- OpenAI Platform API: Service Tiers and Priority Processing
- GPT-6 Astra usage limits guide — frontier model limits, allowance drain, and 5-hour ranges.
- Codex 5-hour limit guide — sliding window mechanics and why Fast mode depletes burst headroom quickly.
- Codex weekly limit guide — 7-day cumulative compute ceiling and weekly reanchoring.
- Codex credits vs. instant resets — purchasing flexible credits to support Fast-mode multipliers.
- why is Codex usage draining so fast — troubleshooting unexpected rapid allowance consumption.