Claude Code

Fast mode in Claude Code: Opus up to 2.5× faster, when is it worth it?

TL;DR

Fast mode keeps Opus and its quality but runs up to 2.5× faster, at a higher price billed to usage credits. Turn it on at session start, since switching mid-session re-bills the whole context uncached, and use it only for interactive work.

You're debugging under pressure, each Opus reply takes tens of seconds, and you wish it were faster. Fast mode is exactly that: the same Opus model, the same quality, up to about 2.5× faster, at a higher price per token. It isn't a different model but an API configuration that prioritises speed. The question is when the extra cost is worth it.

Turning it on and off

/fast                  # toggle (Space to choose, Enter to confirm)
// ~/.claude/settings.json
{ "fastMode": true }
  • When on, a ↯ icon appears next to the prompt. If your current model doesn't support it, Claude Code switches to Opus.
  • Supported on Opus 5.5 (generally available), Opus 5 and Opus 4.8 (research preview). Not on Sonnet or Haiku.
  • By default it stays on across sessions; organisations can require per-session opt-in with fastModePerSessionOptIn.

Pricing and billing

ModelInput / MTokOutput / MTok
Opus 5.5$8$40
Opus 5, Opus 4.8$10$50
  • On Pro, Max, Ultra, Team and Enterprise plans, fast mode draws only from usage credits, not your plan's included usage, even if you have usage left. Usage credits must be turned on first.
  • Team/Enterprise: an Owner must enable fast mode for the organisation.
  • Not available on Amazon Bedrock, Google Cloud or Microsoft Foundry.

The cost trap: turning it on mid-session

Fast mode and standard speed keep separate caches. The first time you turn it on in a conversation, you pay the uncached fast-mode price for the entire existing context. The longer the conversation, the pricier that switch. Every later toggle rereads whatever isn't cached. So: turn it on at the start of a session and don't keep toggling.

When to use it

Use fast modeUse standard speed
Rapid iteration on codeLong autonomous tasks
Live debuggingBatch work, CI/CD
Urgent work with tight deadlinesCost-sensitive workloads

Compared with lowering effort: fast mode keeps quality, lowering only latency and raising price; lowering effort is faster but can reduce quality on hard tasks. You can combine both for simple tasks that need maximum speed.

Risks and caveats

  • Separate rate limits: when you hit them, fast mode falls back to standard speed (the icon greys out) and re-enables after the cooldown.
  • Out of usage credits: Claude Code retries at standard speed and turns fast mode off for that session.
  • Behind an internal gateway: the availability check calls api.anthropic.com directly; if blocked, it may wrongly report fast mode as unavailable.

How to apply it

  • Dev: Turn fast mode on at the start of an urgent debugging session; turn it off when moving to long-running work.
  • Tech Lead: Enable fastModePerSessionOptIn so nobody accidentally leaves fast mode on all week.
  • PM: Track monthly usage-credit spend; fast mode is the line most likely to balloon.

Sources & references

  1. Speed up responses with fast mode — Claude Code Docs