Fast Mode
Fast mode requests a higher-priority capacity tier upstream, which means shorter queuing and faster responses. You can force it on directly at the API key level, with no client-side changes.
For clients without a built-in Fast toggle (such as Codex Desktop), setting the key policy to force Fast on applies automatically.
Costs are subject to model display
Fast mode uses the upstream priority pricing tier. Specific charges are subject to the current display on the Models page; you can also query fast_* pricing for each model via the public API https://tokenflux.dev/api/v1/marketplace/models. Make sure it aligns with your business expectations before enabling it.
How to Set It
- Open the API keys page.
- Create a new key, or click
Editon an existing one. - Find
Fast mode policyand pick an option. - Save.
Policy Options
| Option | Behavior |
|---|---|
Follow request (default) | Passes through whatever the client sent, adding and removing nothing |
Force Fast on | Sends requests as Fast whether or not the client asked for it |
Force Fast off | Strips any Fast flag the client included |
Force Fast off removes only the Fast flag. It leaves other priority settings such as flex, auto, default, and scale untouched.
Scope
The setting applies to all requests made with this key, including regular chat completions, Responses, and Anthropic format.
Changes take effect immediately. Even on long-lived WebSocket connections, the new policy is picked up on the next turn - no reconnect or client restart needed.
When Force On Does Not Apply
Force Fast on has prerequisites. When they are not met it is silently skipped: no error, and the request goes out in regular mode.
- The key is not currently routed to an OpenAI or Anthropic account
- The Anthropic account uses Bedrock, Vertex, or OAuth credentials (only API key credentials are supported)
- The current model does not support the priority tier
- A platform administrator has configured a global Fast filter or block policy (platform policy takes precedence over per-key settings and cannot be bypassed)
Related Links
- Create API Key - how keys are created
- Codex Guide - Codex integration setup
- Billing - billing unit and group multipliers