Execution Speed & Quota Guide

Last verified 10 Sep 2026·Covers Multipliers, CLI Controls, Config & Billing

Codex Fast Mode: Speed, Usage Cost & How to Turn It On

Codex Fast mode is an optional higher-speed serving mode for supported models. The speed gain depends on the model: OpenAI currently says GPT-5.6 Sol can run up to 2.5x faster than Standard, while GPT-5.4 was documented at up to 1.5x faster token velocity. Fast mode consumes usage or credits at a higher rate, with the multiplier also varying by model.

You can enable it in the Codex CLI with /fast on, disable it with /fast off, or inspect your active setting with /fast status. Fast mode changes serving speed rather than model intelligence—OpenAI describes GPT-5.4 Fast mode as the same model and same intelligence, just faster, and says GPT-5.6 Sol Fast mode has no change in intelligence. However, because it burns credits and rolling allowances at an accelerated rate, it is best reserved for interactive pairing rather than unattended background tasks.

Fast Mode vs. Standard Mode: Speed & Credit Multipliers

OpenAI documents specific speedups and credit multiples depending on the active model and billing surface:

Last verified: 10 Sep 2026. From OpenAI documentation and platform pricing.

Model	Expected Speedup	ChatGPT Credit Cost	API Platform Billing	Surface Availability
GPT-5.6 Sol Balanced professional coding & reasoning	Up to 2.5x faster	2.5x Standard	2x standard token pricing	Supported
GPT-5.5 Prior-generation general coding fallback	Faster than Standard; no exact current multiplier published	2.5x Standard	2.5x standard token pricing	Supported
GPT-5.4 Fast agentic scripting & automation	Up to 1.5x faster token velocity	2.0x Standard	2x standard token pricing	Supported
GPT-6 Astra Frontier deep architecture & complex debugging	Faster than Standard; no exact current multiplier published	2.5x Standard (where available)	2x standard ($20/M in, $100/M out)	Supported
GPT-5.3-Codex-Spark Dedicated ultra-velocity Cerebras WSE-3 engine	Exceeds 1,000 tok/s (dedicated hardware)	Separate preview rate limits (does not count toward standard limits)	Not in standard API	Separate Model (Not Fast Mode)

Key Takeaway for Quota Planning

Standard mode executes at 1.0x credit metering. Fast mode boosts serving speed (up to 2.5x on GPT-5.6 Sol and up to 1.5x on GPT-5.4) while charging 2.0x to 2.5x credits depending on the model. If you are coding on a fixed subscription allowance without purchased flexible credits, running in Fast mode burns through your allowance substantially faster per active hour.

What is Codex Fast Mode?

Codex Fast mode is a dedicated runtime setting that prioritizes token throughput and reduces latency across supported OpenAI coding models.

When you send a prompt or initiate an agentic editing task in the Codex CLI, Desktop app, or IDE extensions, the request is typically scheduled on standard inference infrastructure. Under high platform traffic, prompt queuing or heavy token workloads can introduce noticeable latency between turns.

Fast mode routes your session to prioritized compute clusters optimized for higher serving throughput. Depending on the model, speedups range from up to 1.5x token velocity on GPT-5.4 to up to 2.5x faster generation on GPT-5.6 Sol (while OpenAI notes GPT-5.5 and Astra run faster than Standard without publishing exact multipliers). In the OpenAI API, this capability was previously referred to as priority processing and is configured using the service_tier: "fast" parameter.

CLI Controls: How to Turn Fast Mode On, Off, and Check Status

In any active Codex CLI terminal session, you can toggle or inspect Fast mode dynamically using slash commands:

1. Enable Fast Mode

Activates high-throughput inference for the active session. Only works when using a model that supports Fast mode.

/fast on

2. Disable Fast Mode

Returns the active session to standard processing speed and baseline 1.0x credit consumption.

/fast off

3. Check Active Status

Prints whether Fast mode is currently enabled, disabled, or unavailable for your loaded model.

/fast status

Model Compatibility Warning

If your active model does not support a fast tier (such as certain legacy models or specialized preview endpoints), running /fast on will notify you that Fast mode is unavailable for that model.

Persistent Configuration: Setting Fast Mode in config.toml

Running /fast on only persists for the duration of that specific CLI session. To make Fast mode your default across all future sessions, configure it in your user configuration file:

~/.codex/config.toml

# Preferred service tier for turns and threads
service_tier = "fast"

[features]
# Enables Fast mode toggle and model-catalog fast tier
fast_mode = true

OpenAI documents two key configuration fields:

service_tier = "fast": Sets the default routing tier for new requests to Fast mode.
[features] fast_mode = true: Enables Fast-tier selection and slash command features in the CLI and interactive TUI.

Enterprise & Workspace Overrides: In ChatGPT Business and Enterprise workspaces, admins can manage Fast mode availability centrally under Workspace Settings > Models or enforce it via requirements.toml. Local settings in config.toml will yield to organizational policies.

Does Codex Fast Mode Lower Code Quality or Reasoning?

No. Fast mode changes the serving speed, not the model's intelligence.

A common concern is that Fast mode functions like a smaller model or reduces reasoning quality. OpenAI describes GPT-5.4 Fast mode as the same model and same intelligence, just faster, and explicitly confirms that GPT-5.6 Sol Fast mode has no change in intelligence.

If you have configured High reasoning effort on GPT-5.6 Sol or GPT-6 Astra, the model executes its full reasoning process before streaming responses. Serving acceleration originates from prioritized infrastructure scheduling and dedicated capacity, not from cutting reasoning corners or reducing intelligence.

Does Codex Fast Mode Use More Quota? (5-Hour & Weekly Impact)

Yes, Codex Fast mode exhausts both your rolling 5-hour limit and your 7-day weekly allowance substantially faster.

5-Hour Rolling Window Impact

Because generation completes significantly faster, you take more actions in less time. If you run consecutive prompts in Fast mode, you can easily exhaust your entire 5-hour burst allowance in a fraction of normal working time. Learn more about sliding mechanics in our Codex 5-hour limit guide.

Weekly Limit & Credit Multipliers

For accounts operating on credit metering or shared token capacity, Fast mode applies a 2.0x to 2.5x multiplier to consumption. Each turn drains more than double the credit volume of a Standard turn. Review our guide on Codex weekly limit mechanics.

If you notice unexpected quota depletion, verify whether Fast mode was accidentally left enabled in config.toml using our troubleshooting guide for rapid quota drain.

GPT-6 Astra Fast Mode

GPT-6 Astra supports Fast mode across eligible ChatGPT plans and the OpenAI Platform API, delivering faster generation than Standard processing (OpenAI has not published a specific standalone speed multiplier for Astra).

Because Astra is OpenAI's heaviest frontier reasoning model, standard generation can take several seconds to stream extensive multi-file refactors. Fast mode noticeably cuts this latency. However, Astra Fast mode consumes credits at a 2.5x Standard multiplier where available.

In the developer API, Astra Fast mode routes through priority infrastructure at $20/M input tokens and $100/M output tokens (2x the standard $10/M input and $50/M output rates). For comprehensive allowance ranges, see our GPT-6 Astra usage limits guide.

Fast Mode vs. Codex-Spark: Don't Confuse Them

A frequent point of confusion among developers is treating “Fast mode” and “Codex-Spark” as interchangeable terms. They are completely different features:

Codex Fast Mode

A Speed & Priority Tier

Applies to general models (GPT-5.4, GPT-5.5, GPT-5.6 Sol, GPT-6 Astra).
Delivers model-dependent speed increases (up to 2.5x on GPT-5.6 Sol, up to 1.5x on GPT-5.4).
Consumes regular subscription credits at 2x to 2.5x multiples.
Toggled via /fast on or service_tier.

GPT-5.3-Codex-Spark

A Separate Dedicated Model

Separate research-preview model served on Cerebras Wafer Scale Engine 3 hardware.
Can exceed 1,000 tokens per second for high-velocity coding.
Operates with separate rate limits whose usage does not count toward standard rate limits during preview.
Available to eligible subscribers during the research preview.

For full details on Spark's separate rate-limit bucket and hardware architecture, read our GPT-5.3-Codex-Spark limits guide.

ChatGPT Sign-in vs. API-Key Usage: Billing Differences

How Fast mode is billed depends entirely on how your Codex CLI is authenticated.

ChatGPT Subscription Authentication

When you log in via codex login with a ChatGPT Plus, Pro, or Business subscription, Fast mode draws against your included monthly plan allowance or purchased flexible credit balance. Turns consume credits at 2.0x or 2.5x Standard.

OpenAI Platform API Key Authentication

If you authenticate the CLI with an OPENAI_API_KEY, requests route directly to developer API billing. Fast mode activates service_tier: "fast", which bills directly per million tokens at model-specific Fast rates (such as 2x Standard on GPT-5.4 and GPT-6 Astra, and 2.5x Standard on GPT-5.5 at $12.50/M input and $75/M output), bypassing 5-hour and weekly subscription caps.

Compare subscription economics and token rates in our complete Codex pricing comparison guide and credits vs. resets guide.

When is Codex Fast Mode Actually Worth Using?

Because Fast mode costs 2x to 2.5x more credits, using it selectively maximizes developer productivity without prematurely running out of quota:

When to Use Fast Mode

Interactive Pair Programming: When you are sitting at the keyboard waiting for immediate feedback to stay in flow.
Live Bug Triage: During production incidents or crunch debugging where seconds saved directly matter.
Presentations & Demos: When streaming code live to colleagues or clients without awkward waiting pauses.
Generous Credit Balances: When your account has ample flexible credits or an employer-funded API tier.

When to Stick with Standard Mode

Unattended Background Tasks: Batch test generation, broad repo sweeps, or multi-step agent loops.
Large File Summaries: Reading hundreds of files where input context dominates and output speed is irrelevant.
Tight Allowance Budgets: When working on a Plus or Business Standard tier without purchased credit buffers.
Friday Afternoon Sprints: When depleting your weekly allowance early would freeze access for the weekend.

Frequently Asked Questions About Codex Fast Mode

What is Codex Fast mode?

How much does Codex Fast mode cost in credits?

In ChatGPT subscription plans with credit metering, Fast mode consumes 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. On the OpenAI Platform API, Fast mode is billed using model-specific rates—such as 2x Standard for GPT-5.4 and GPT-6 Astra, and 2.5x Standard for GPT-5.5 ($12.50/M input, $75/M output).

How do I enable Codex Fast mode in the CLI?

In the Codex CLI, type /fast on to enable Fast mode for your active session, /fast off to return to standard speed, and /fast status to check your current mode. To make it persistent across sessions, add service_tier = "fast" to your ~/.codex/config.toml.

Does Codex Fast mode produce lower-quality code or reduce reasoning?

No. Fast mode changes serving speed, not the model's intelligence. OpenAI describes GPT-5.4 Fast mode as the same model and same intelligence, just faster, and says GPT-5.6 Sol Fast mode has no change in intelligence.

Is Codex-Spark the same as Codex Fast mode?

No. GPT-5.3-Codex-Spark is a separate research-preview model served on Cerebras Wafer Scale Engine 3 hardware that can exceed 1,000 tokens per second, with separate rate limits whose usage does not count toward standard rate limits. Fast mode is an optional speed tier for general models that burns standard credits or allowance.

Authoritative Sources

The facts, commands, and multipliers on this page are verified against official OpenAI documentation:

Limits, Speed & Quota Management

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.