Is standard-speed Opus 4.8 retired too?
Per the third-party write-up, the notice does not mention retiring standard-speed Opus 4.8; what is removed is the fast mode option.
That is the write-up's account, so to confirm the model's own lifecycle, check Anthropic's official model deprecation page.
I only use /fast in Claude Code. Do I need to do anything?
Per that write-up, Claude Code has defaulted fast mode to Opus 5.5 since v2.1.280, so ordinary installs may already be fine.
If you have settings that pin the model version, check whether they still point at Opus 4.8.
Is the Opus 5.5 fast price reliable?
The third-party write-up says current documentation and the launch page list $8 per million input and $40 output, but an earlier report had flagged it as unconfirmed.
Go by Anthropic's official pricing page, then check your own bill against the usage fields.
How do I confirm I actually got fast?
Per that write-up, a successful response should show fast in usage.speed. Putting that field into your tests or log checks is more reliable than eyeballing latency.
If it shows standard, the request did not take fast, and your cost and speed expectations both need recalculating.
The Claude Code v2.1.316 changelog (October 31, 2026) says fast mode for Opus 4.8 is deprecated as of October 30 and removed on November 30, after which requests return an error instead of falling back to standard-speed Opus 4.8. With less than a month to the removal date, the people most likely to get hurt are those who hard-coded the old model ID and a fast flag in scripts. What follows relies on the changelog's description; the pricing and migration details come from a third-party write-up by Developers Digest, and I did not read the official documentation directly.
Deprecation is October 30 and removal is November 30. Per the third-party write-up, after removal a fast mode request on Opus 4.8 returns an error and does not fall back to standard Opus 4.8. The notice does not mention retiring standard-speed Opus 4.8, so what is affected is the fast mode option, not the whole model. That matters, because no fallback means what used to become quietly slower will become an outright failure.
That write-up lists Opus 4.8 fast mode at $10 per million input tokens and $50 per million output tokens, and Opus 5.5 fast mode at $8 and $40, 20% less on both. The author gives no standard price for Opus 4.8; his inference is that the fast premium is about double, but that is inference, not the article's claim. The migration is to change the model ID from claude-opus-4-8 to claude-opus-5-5; Opus 5.5 fast mode needs no beta header, whereas Opus 4.8 requests use fast-mode-2026-02-01. On success the response's usage.speed should read fast.
The article lists the affected places: /fast in Claude Code (per the quoted documentation, if the current model does not support fast mode, it switches to Opus), speed: "fast" in API requests, and scripts, CI or harnesses that pair the old model ID with the fast flag. It also says Claude Code has defaulted fast mode to Opus 5.5 since v2.1.280, so many ordinary installs may already be fine. CLAUDE_CODE_SUBAGENT_EFFORT, added in the same release, is unrelated to this change.
This write-up is third-party, and I can only see its account of the changelog, documentation and migration guide. Opus 5.5 fast pricing was flagged as unconfirmed in an earlier report; this article says the current docs and launch page now list $8 and $40. Its Python example passes speed="fast" as a parameter, but the author does not show that call verified with the SDK. For the move to Opus 5.5, the article also lists other behavior differences, such as thinking on by default and not disableable, and the removal of sampling parameters like Temperature; compare against the official migration guide yourself before switching.
First find who still uses the old combination: search your scripts, CI settings and frameworks for claude-opus-4-8 paired with a fast flag. If you find it, switch to Opus 5.5 before November 30 and verify with usage.speed that you really got fast. At the list prices the third party gives, fast saves $2 per million input tokens and $10 per million output tokens, which is worth computing once for large batch work; but because changing models also changes behavior, compare output quality and total spend in a test environment before moving production traffic.