Claude Opus 4.8 fast mode ends on November 30

On October 30, Anthropic announced in the Claude Release Notes that Claude Opus 4.8 fast mode is deprecated and will be removed on November 30, 2026. After that date, requests to claude-opus-4-8 with the speed: "fast" parameter will return an error, and Anthropic recommends moving to Claude Opus 5.5.

What exactly changes in Claude Opus 4.8 fast mode?

Fast mode is a faster inference configuration of the same model: up to 2.5x more output tokens per second, with no change to the model’s intelligence (I wrote more about it when covering fast mode for Opus 5.5). On Opus 4.8 it only ever ran as a research preview, meaning with the fast-mode-2026-02-01 beta header and after being granted access.

According to the fast mode documentation:

1. Until November 30, fast mode on Opus 4.8 works as before.
2. After November 30, a request with speed: "fast" to claude-opus-4-8 will return an error rather than quietly falling back to standard speed.
3. The deprecation covers fast mode only. The announcement says nothing about the Opus 4.8 model itself.

Point 2 matters most for anyone with fast mode wired into production. If we don’t switch models or handle that error, those requests will start failing after November 30.

Where does fast mode stay?

The list of models with fast mode shrinks to two:

Model	Fast mode status	Price (input / output per million tokens)
Claude Opus 5.5	available, no beta header	$8 / $40
Claude Opus 5	research preview	$10 / $50
Claude Opus 4.8	research preview, until November 30	$10 / $50

Prices come from the fast mode documentation. Interestingly, moving to Opus 5.5 cuts the fast mode rate by 20%, and Opus 5.5 no longer needs a waitlist. Organizations in the Start, Build and Scale tiers get a default fast mode rate limit right away.

How do you migrate from Opus 4.8 to Opus 5.5?

Changing the model ID is only the start. The migration guide lists several things that are new for code written against Opus 4.8:

1. Thinking is always on. On Opus 4.8, a request without a thinking field ran without thinking. On Opus 5.5, adaptive thinking is always on, and thinking: {"type": "disabled"} returns a 400 error. Thinking tokens count as output, so it’s worth reviewing max_tokens and costs.
2. Thinking depth is set with the effort parameter. The default level on Opus 5.5 is medium, versus high on Opus 4.8. For xhigh and max, the documentation recommends max_tokens of 64k or more.
3. No more forced tool use. A tool_choice of type any or tool returns an error. What’s left is auto (or none) plus strict tool use or structured outputs.
4. No temperature, top_p, top_k or prefill. These parameters have to be removed from requests.
5. Priority Tier doesn’t work on Opus 5.5. If you have such a commitment, you need to plan capacity separately (and fast mode doesn’t combine with Priority Tier anyway).

There’s good news too: the minimum prompt length for caching drops from 1024 to 512 tokens, so shorter prompts will start getting cached as well, with no code changes.

The documentation also suggests a shortcut: in Claude Code, you can hand off the migration with the command /claude-api migrate this project to claude-opus-5-5 ;)

What should you watch out for?

First, error handling. With fast mode, it’s worth having a fallback path to standard speed, because even on Opus 5.5 exceeding the fast mode limit returns a 429 or 529 error. Just remember that switching speeds means a prompt cache miss.

Second, costs after the migration. A lower per-token rate doesn’t have to mean a lower bill if tasks that used to run without thinking now generate thinking tokens. The documentation explicitly advises re-measuring cost and latency at your chosen effort level, and the new Claude API cost analytics, which also shows fast mode usage, can help with that.

FAQ

When will Claude Opus 4.8 fast mode stop working?

On November 30, 2026. After that date, requests to claude-opus-4-8 with speed: "fast" will return an error.

Is Claude Opus 4.8 itself being retired too?

The October 30 announcement covers fast mode only. Regular requests to Opus 4.8 (without speed: "fast") aren’t affected by it.

How much does fast mode cost on Opus 5.5?

$8 per million input tokens and $40 per million output tokens, which is less than the $10/$50 on Opus 4.8 and Opus 5.

Does fast mode on Opus 5.5 need a beta header?

No. speed: "fast" in the request is enough. The fast-mode-2026-02-01 header is only needed for Opus 5 and Opus 4.8.