Fast mode
An option some providers offer to serve a model with faster output generation, usually at a higher price.
Why it matters. Useful when response speed matters more than cost, such as interactive coding or live agents.
How it works
Fast mode is an option some providers offer that generates a model's output at higher speed, in tokens per second, than its standard mode. It is typically the same model served on a faster configuration, not a smaller or different model.
Availability, price and the speed gain vary by provider and model, and it may be offered only as a preview or to some accounts. It is usually priced at a premium over standard rates and turned on per request. This site's model pages show whether a model has a fast mode when the maker documents it.
Use it on routes where people wait on long outputs. It does not reduce the work the model does, so lowering effort or switching to a smaller model remain separate levers that also cut cost.
Worked example
Illustrative: a 3,000-token answer at 60 tokens per second takes 50 s; at 150 tokens per second it takes 20 s. If fast mode doubles a $15 per 1M output rate, that answer goes from $0.045 to $0.09.
Common mistakes
- Turning it on for background jobs where nobody is waiting on the output.
- Assuming fast mode is a cheaper or smaller model; it usually costs more.
- Building on it without checking whether it is a preview with limited availability.
See it on a real card: Specifications on Claude Opus 5.5.