Fast mode

Fast mode buys latency with money. Same model, faster output, premium rate.

Where it works, and where it does not

Model	Fast mode
Claude Opus 5	Yes
Claude Opus 4.8	Yes
Claude Opus 4.7	No — speed: "fast" returns an error
Claude Opus 4.6	Requests run at standard speed and are billed at standard rates

That third row is the trap: on Opus 4.7 you get a clear error, but on Opus 4.6 the request quietly succeeds at normal speed. No failure, no speed-up.

First-party Claude API only. Not available on Claude Platform on AWS, Amazon Bedrock, Google Cloud or Microsoft Foundry.

Pricing

Rate	Fast mode	Standard
Input	$10 / MTok	$5 / MTok
Output	$50 / MTok	$25 / MTok

Double, and the fast rate applies across the full context window, including requests above 200K input tokens.

Multipliers stack on top: prompt caching rates apply to the fast base price, as does the inference_geo residency multiplier. Fast mode is not available with the Batch API — which makes sense, since batching trades latency for cost in the opposite direction.

In Claude Code

Toggled with /fast. It uses Claude Opus with faster output; it does not switch you to a smaller model. A frequent misconception worth stating plainly.