Playbooks · Coding tasks
Move fast-mode work off Opus 4.8 in time
Written November 2, 2026
Fast mode on Opus 4.8 is deprecated and removed November 30. After removal, fast-mode requests error and do not fall back. Migrate to Opus 5.5.
What changes on November 30
The Claude Code changelog for version 2.1.316 states: "Deprecated fast mode for Opus 4.8 as of October 30, 2026; it will be removed on November 30, 2026. After removal, fast mode requests on Opus 4.8 return an error and do not fall back to standard Opus 4.8. Migrate to Opus 5.5 to keep the speedup." The Claude Platform release notes say the same for API requests: after removal, requests to claude-opus-4-8 with speed: "fast" will return an error, and the migration is to Claude Opus 5.5. This is an error, not a slowdown. Anything that still asks for fast mode on Opus 4.8 after November 30 breaks instead of running at standard speed.
Fast mode is a high-speed configuration of Claude Opus, not a different model: Anthropic's fast-mode documentation describes it as making the model up to 2.5 times faster in output tokens per second at a higher cost per token, with the same quality and capabilities. It is supported on Opus 5.5, Opus 5, and Opus 4.8 only. Opus 5.5 is the fast-mode default in Claude Code version 2.1.280 and later, and fast mode on Opus 5.5 needs no beta header.
Before you start
Fast mode can be turned on in more than one place: the /fast toggle in a session, the fastMode setting in a settings file, an environment variable that disables it, and — for API use — the speed: "fast" request field. Check all of them. A session toggle you set months ago and forgot is still a fast-mode use, and so is a script that pins the model and the speed together.
Steps
- Find every fast-mode use on Opus 4.8. Check your Claude Code settings files for
fastMode, and check whether the model in use with it is Opus 4.8. Search your scripts, settings, and automation forfastMode,Opus 4.8, andclaude-opus-4-8, and search API code forspeed: "fast"sent with the Opus 4.8 model. Include non-interactive runs: in non-interactive mode,/fastworks only in a session launched with fast mode in its--settingsvalue, so check those launch commands too. Write down each use: where it lives, what it does, and whether it must keep fast mode or only happens to have it on. - Move each use to Opus 5.5. Change the model for each fast-mode use from Opus 4.8 to Opus 5.5, keeping fast mode on where the work still needs the speed. Do not move the work to standard speed on Opus 4.8 and call it migrated: after November 30, a fast-mode request on Opus 4.8 errors even if standard Opus 4.8 still exists. If a use does not need the speed, turn fast mode off for that use instead — fast mode costs more per token, at $8 per million input tokens and $40 per million output tokens on Opus 5.5 — but turn it off explicitly rather than leaving a fast-mode request pointed at Opus 4.8.
- Test one fast-mode request on Opus 5.5 before November 30. In Claude Code, run
/fastand confirm the "Fast mode ON" message and the fast-mode icon beside the prompt, then run one request on Opus 5.5 and confirm it completes. For API use, send one request to Opus 5.5 withspeed: "fast"and confirm it succeeds. If the request fails, the failure is in front of you while Opus 4.8 fast mode still works — not after it errors. - Check the remaining surfaces. Fast mode behaves per surface: the VS Code extension offers a Toggle fast mode command when the selected model supports it, and fast mode needs usage credits turned on for subscription plans. Check each surface you actually use — terminal, VS Code extension, cloud sessions, API — for a fast-mode setting still naming Opus 4.8. If your organization disables fast mode entirely,
CLAUDE_CODE_DISABLE_FAST_MODE=1is the documented way to do it; that is a different decision from this migration, and it does not substitute for moving Opus 4.8 uses that keep fast mode.
What can go wrong
- Expecting a fallback. There is none. After removal, fast-mode requests on Opus 4.8 return an error and do not fall back to standard Opus 4.8. Scripts and automation that assume a graceful slowdown will fail outright.
- Moving the model but keeping the old request. An API call that still names
claude-opus-4-8withspeed: "fast"errors after November 30, whatever the rest of the code does. Both the model and the speed in the request must move together. - Missing a non-interactive launch. A scheduled or scripted run that launches with fast mode in its
--settingsvalue and Opus 4.8 as its model is a fast-mode use, even though no one toggles/fastfor it. - Assuming other models are affected. The removal covers fast mode on Opus 4.8 only. Fast mode on Opus 5.5 and Opus 5 is not removed by this change, and fast mode never applied to Sonnet, Haiku, or other models.
- Paying for speed the work does not need. Fast mode costs more per token than standard speed on the same model. Keep it where response speed matters, and turn it off where it does not — deliberately, per use, not by leaving it on everywhere.
Limits
- This page covers fast mode on Opus 4.8 only, as the changelog entry states it. It makes no claim about fast mode on any other model beyond the supported-model list in Anthropic's documentation.
- The "up to 2.5 times faster" figure is Anthropic's description of fast mode's output rate. It is not a measured result for your work; the test in step 3 confirms only that the request runs, not how much faster it is.
- Fast mode availability depends on your account: subscription plans need usage credits turned on, and Team and Enterprise organizations need an Owner to enable it. If fast mode is unavailable on your account, that is a separate check from this migration.
Related playbooks
- Claude Code 2.1.322: Vertex fast mode, three regions
- Move off Claude Sonnet 4.5 before it retires
- Fable 5.2 vs Opus 5.5 on real coding work
- Move Cursor work before the OpenAI shutoff
Sources
- https://code.claude.com/docs/en/changelog
- https://code.claude.com/docs/en/fast-mode
- https://platform.claude.com/docs/en/release-notes/overview
Back to: Coding playbooks hub · Also: Haiku 5.5 is 90% cheaper only under 100K tokens: price the tier jump and the tokenizer before you route background work to it · Pick the cheapest model for the job · Measure what Claude Code's background-agent summaries cost you, and decide what to turn off