Featured Claude Code Models Fast Mode Opus 5.5 Claude API Migration

Opus 4.8 Fast Mode Ends Nov 30: How to Migrate to Opus 5.5

Fast mode for Claude Opus 4.8 is removed on November 30, 2026. What breaks, who is affected, and how to move API calls and Claude Code to Opus 5.5.

• 8 min read • By Claude World

Anthropic deprecated fast mode for Claude Opus 4.8 on October 30, 2026, and removes it on November 30, 2026. From then on, fast-mode requests to Opus 4.8 return an error, while Opus 4.8 at standard speed keeps working. To keep the speedup, move to Claude Opus 5.5, whose fast mode costs less per token and no longer needs a beta header.

Checked November 4, 2026 against the Claude Platform and Claude Code documentation and the Claude Code v2.1.316 changelog. Sources are linked at the end.

What Changes, and When

  • October 30, 2026: fast mode for Opus 4.8 is deprecated, per the Claude Platform release notes and Claude Code v2.1.316.
  • November 30, 2026: it is removed. Requests to claude-opus-4-8 with speed: "fast" then return an error, and Claude Code’s changelog adds that they “do not fall back to standard Opus 4.8.”
  • Unchanged: Opus 4.8 at standard speed, which Anthropic’s model deprecations page lists as Active with a tentative retirement date “Not sooner than May 28, 2027,” and fast mode on Opus 5.5 and Opus 5.

Fast mode has been removed from Opus models before, with different results:

ModelFast mode deprecatedFast mode removedFast requests after removal
Opus 4.6May 28, 2026June 29, 2026Run at standard speed and standard rates, with no error
Opus 4.7June 25, 2026July 24, 2026Return an error; no fallback to standard speed
Opus 4.8October 30, 2026November 30, 2026Return an error; Claude Code does not fall back to standard Opus 4.8

Opus 4.8 follows the Opus 4.7 pattern, so treat November 30 as a hard cutoff, not a silent downgrade.

Who Is Affected

Two groups need to act:

  • Claude API callers that send model: "claude-opus-4-8" with speed: "fast" to the Messages API, or run Managed Agents whose model object is {"id": "claude-opus-4-8", "speed": "fast"}. Fast mode on Opus 4.8 is a research preview on the Claude API only and needs the fast-mode-2026-02-01 beta header.
  • Claude Code users whose session model is Opus 4.8 with fast mode on. Claude Code’s docs warn: “If Opus 4.8 is still your model with fast mode on after that date, the API rejects your requests.”

You are not affected if you run Opus 4.8 at standard speed, already use fast mode on Opus 5.5 (Claude Code’s fast-mode default since v2.1.280), or run Claude Code on Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry, or Claude Platform on AWS, where Claude Code has no fast mode.

Migrating API Calls

1. Change the model, keep speed: "fast"

Replace claude-opus-4-8 with claude-opus-5-5. Fast mode on Opus 5.5 left research preview on October 15, 2026, and the release notes say: “Requests to Claude Opus 5.5 no longer require the fast-mode-2026-02-01 beta header, and requests that still send it keep working.” Fast mode on Opus 5 still needs the header.

 curl https://api.anthropic.com/v1/messages \
   -H "x-api-key: $ANTHROPIC_API_KEY" \
   -H "anthropic-version: 2026-10-21" \
-  -H "anthropic-beta: fast-mode-2026-02-01" \
   -H "content-type: application/json" \
   -d '{
-    "model": "claude-opus-4-8",
+    "model": "claude-opus-5-5",
     "max_tokens": 1024,
     "speed": "fast",
     "messages": [{"role": "user", "content": "Hello"}]
   }'

Dropping the header line is optional. The SDK equivalents, following the documentation’s examples:

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=1024,
    speed="fast",
    messages=[{"role": "user", "content": "Hello"}],
)

print(response.usage.speed)  # "fast" or "standard"
import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic();

const response = await client.messages.create({
  model: "claude-opus-5-5",
  max_tokens: 1024,
  speed: "fast",
  messages: [{ role: "user", content: "Hello" }]
});

console.log(response.usage.speed); // "fast" or "standard"

For Managed Agents, set the agent’s model object to {"id": "claude-opus-5-5", "speed": "fast"}; the migration guide says Managed Agents need no changes beyond the model name.

2. Confirm the speed with usage.speed

Every response’s usage.speed reports "fast" or "standard", so log it during the cutover. A successful fast request reports "fast"; a model without fast mode, or an exceeded fast-mode rate limit or capacity (429 or 529), returns an error instead.

3. Check price and rate limits

ModelStandard (input / output per MTok)Fast mode (input / output per MTok)
Opus 5.5$4 / $20$8 / $40
Opus 5$5 / $25$10 / $50
Opus 4.8$5 / $25$10 / $50

Per token, fast mode on Opus 5.5 costs 20% less than on Opus 4.8. Fast-mode rates cover the full context window, including requests over 200k input tokens; prompt caching and data residency multipliers apply on top; and fast mode isn’t available with the Batch API or a Priority Tier commitment.

Fast mode has its own rate limit, separate from standard Opus limits. Opus 5.5 gets a default one in the Start, Build, and Scale usage tiers, with no access request; Evaluation-tier organizations get it once they move up. See yours on the Console’s Rate limits page or in the anthropic-fast-* response headers. Going over returns a 429 with retry-after, which the SDKs retry up to twice by default.

4. If you fall back to standard speed, expect a cache miss

The documented fallback sets max_retries to 0 on the fast request, catches the rate-limit error, and retries without speed: "fast". Note the docs’ warning: “Falling back from fast to standard speed results in a prompt cache miss. Requests at different speeds do not share cached prefixes.” The pattern handles rate limits; it does not replace changing the model ID.

5. Read the Opus 4.8 section of the migration guide

Opus 5.5 is a different model, not a faster Opus 4.8, so run your evals before moving production traffic. The Opus 5.5 migration guide has a section for code written against Opus 4.8. Among its points: on the Claude API, Opus 5.5 returns a 400 error for thinking: {"type": "disabled"}, forced tool choice (tool_choice of any or tool), and the computer_20251124 tool, all of which Opus 4.8 accepts; thinking runs even on requests that omit it, so revisit max_tokens and read content blocks by type; the default effort is medium, where Opus 4.8 defaults to high; and Priority Tier is not supported. The guide also suggests letting Claude Code make the edits with /claude-api migrate this project to claude-opus-5-5.

Migrating Claude Code

The Claude Code docs offer two ways out. Pick one before November 30:

  • Keep fast mode: switch to Opus 5.5. Run /model claude-opus-5-5, which requires Claude Code v2.1.280 or later. Fast mode stays on after the switch, and typing a model name after /model saves it as your default for new sessions. /model opus also selects Opus 5.5 on the Anthropic API, unless ANTHROPIC_DEFAULT_OPUS_MODEL pins the alias elsewhere.
  • Keep Opus 4.8: turn fast mode off. Run /fast off to stay on Opus 4.8 at standard speed.

Switch at the start of a session if you can. Each model has its own prompt cache, so the first request after /model reads the whole conversation with no cache hits, and fast mode and standard speed keep separate caches too.

Check where you stand

  • /status shows your Claude Code version, model, and account.
  • /fast shows whether fast mode is on, and a ↯ icon sits next to the prompt while it is.
  • claude --version shows your version, and claude update upgrades it.

If a new session still starts on Opus 4.8 after you switch, something with higher priority sets the model: --model, ANTHROPIC_MODEL, a model value in project or managed settings, or an organization default configured to override user choices. Sessions resumed with claude --resume or --continue usually keep the model they were using, too.

Billing on subscription plans

On Pro, Max, Ultra, Team, and Enterprise plans, fast mode bills to usage credits only, even when you have plan usage left, and it isn’t included in your plan’s rate limits. Switching from Opus 4.8 to Opus 5.5 lowers that rate from $10 / $50 to $8 / $40 per MTok. If you sign in with a Claude Console account, fast mode bills per token with the rest of your organization’s API usage.

What admins should check

  • Model allowlist. If availableModels excludes Opus 5.5, users can’t take the keep-fast-mode path: /model rejects the switch, and /fast refuses to move a session to it with “is not in your organization’s allowed models.” Allow claude-opus-5-5.
  • Pinned models. A model value in managed settings, or an organization default set to override user selection, reapplies at every launch. Update any that still name Opus 4.8.
  • Enablement. Fast mode is disabled by default for Team and Enterprise organizations until an Owner enables it under Organization settings > Claude Code (Console admins use Claude Code preferences), and a member with billing access must turn on usage credits under Organization settings > Usage.
  • Policy keys. Managed settings with fastMode: false make /fast report that your organization has disabled fast mode; fastModePerSessionOptIn: true starts every session with fast mode off; CLAUDE_CODE_DISABLE_FAST_MODE=1 disables it entirely.
  • Gateways. The fast-mode availability check calls api.anthropic.com directly, not through ANTHROPIC_BASE_URL. If /fast says fast mode is unavailable or disabled behind an LLM gateway although your organization enabled it, see CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS and CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK in the fast mode docs.

Checklist Before November 30

  • Search code, configs, scripts, and agent definitions for claude-opus-4-8 combined with speed: "fast" or fastMode.
  • API: switch to claude-opus-5-5 and keep speed: "fast"; the beta header is optional for Opus 5.5.
  • Work through the migration guide’s Opus 4.8 section and run your evals on Opus 5.5.
  • Confirm your organization is in the Start, Build, or Scale usage tier.
  • Log usage.speed, then re-baseline cost and latency.
  • Claude Code: update to v2.1.280 or later, check /status and /fast, then run /model claude-opus-5-5 or /fast off.
  • Clear old pins: --model, ANTHROPIC_MODEL, model in settings files, and organization defaults.
  • Admins: allow Opus 5.5 in availableModels, and confirm fast mode is enabled and usage credits are on.

FAQ

Will Claude Opus 4.8 stop working on November 30?

No. Only its fast mode goes away. Standard-speed requests keep working, and claude-opus-4-8 is listed as Active with retirement not sooner than May 28, 2027.

Will fast requests to Opus 4.8 fall back to standard speed automatically?

No. They return an error and, per Claude Code’s changelog, “do not fall back to standard Opus 4.8.” Opus 4.6 was the exception: after its removal, fast requests ran at standard speed without an error.

Is fast mode on Opus 5.5 more expensive than on Opus 4.8?

Per token it is cheaper: $8 / $40 per MTok against $10 / $50, though still double Opus 5.5’s standard $4 / $20. Opus 5.5 also runs thinking on requests that omit the thinking field, and thinking bills as output, so measure per-request cost too.

Do I still need the fast-mode-2026-02-01 beta header?

Not for Opus 5.5, and requests that still send it keep working. Fast mode on Opus 5 remains a research preview and still needs it.

Can I use fast mode on Sonnet, Haiku, or Fable instead?

No. Fast mode runs only on Opus 5.5, Opus 5, and Opus 4.8. Opus 5 remains a research-preview option at $10 / $50 (on the API, access goes through your account manager or the waitlist), but the migration path Anthropic names is Opus 5.5.

Sources