Two Claude cutoffs land on November 30: Opus 4.8 fast mode and Sonnet 4.5

Published November 1, 20265 min read

Claude Code 2.1.316 announced the fast mode change in its changelog: “Deprecated fast mode for Opus 4.8 as of October 30, 2026; it will be removed on November 30, 2026.” The API fast mode docs say the same, and spell out what happens next: requests to claude-opus-4-8 with speed: "fast" “will return an error.”

The Sonnet 4.5 retirement is older news that's easy to miss. Anthropic's deprecations page listed it on September 30, with claude-sonnet-4-5-20250929 retiring November 30 and claude-sonnet-5-5 as the replacement. So one date now covers two different breakages. This page is based on Anthropic's docs and changelog; we didn't benchmark the replacement models for this piece.

What changes on November 30

The middle row matters. Opus 4.8 itself isn't going away this year, only its fast configuration. A pipeline that runs Opus 4.8 at normal speed keeps working.

Fast mode on Opus 5.5 costs less

Fast mode is the same model with a different serving configuration. Claude Code's fast mode docs describe it as “up to 2.5x faster at a higher cost per token.” On every supported model, it's priced at twice the standard rate, per Anthropic's pricing page:

So the forced move is also a 20% price cut on fast mode. Take a session that sends 1M uncached input tokens and gets back 100K output tokens. On Opus 4.8 fast that's $10 + $5 = $15. On Opus 5.5 fast it's $8 + $4 = $12. The same session at standard Opus 5.5 speed is $6. Caching lowers all three, but the ratios hold, since cache multipliers stack on top of fast mode pricing.

Opus 5.5 fast mode is also easier to get on the API. The docs list Opus 4.8 and Opus 5 fast mode as a research preview that needs the fast-mode-2026-02-01 beta header and provisioned access. Opus 5.5 needs no beta header, and organizations on the Start, Build, and Scale tiers get a default fast mode rate limit without asking.

If you use Claude Code

Most Claude Code users are already fine. Opus 5.5 has been the fast mode default since Claude Code v2.1.280, per the fast mode docs. You're affected only if you've pinned Opus 4.8 and then turned on /fast. Check these places:

• Your model setting. Run /model and /fast to see what you're on. Look for claude-opus-4-8 in ~/.claude/settings.json, project .claude/settings.json, and the ANTHROPIC_MODEL environment variable.
• Headless jobs. Search CI configs and scripts for claude -p calls that pass both an Opus 4.8 model and "fastMode": true in --settings.
• Admin allowlists. If your organization's availableModels list allows Opus 4.8 but not Opus 5.5, /fast will be refused once 4.8 fast mode goes. Add Opus 5.5 before the 30th.

One cost detail from the docs: the first time you turn on fast mode in a conversation, you pay the full uncached fast mode input price for the whole context. Fast and standard speed also keep separate caches. Switching models mid-session reads the context uncached again too, so make the change at the start of a session, not halfway through a long one.

If you call the API directly

Search your code for three strings. claude-sonnet-4-5 finds Sonnet 4.5 calls. fast-mode-2026-02-01 finds fast mode calls on the older models. claude-opus-4-8 finds every Opus 4.8 call, and you need to check each match for a fast speed setting by hand.

grep -rnE "claude-sonnet-4-5|fast-mode-2026-02-01|claude-opus-4-8" .

Don't filter the Opus 4.8 matches down to lines that also mention speed. The model and speed: "fast" are often on separate lines, and the speed can be a default on a shared client or wrapper.

To see what's really running, check the usage.speed field on responses, which is "fast" or "standard". In the Console, the Usage and Cost pages can group by Speed (Research Preview), which breaks out fast mode spend by date.

Watch your error handling. The API treats a fast mode request it can't serve as an error, not a silent downgrade. A retry loop that retries every failure will just burn retries on a request that can never succeed. Claude Code handles rate limits differently: when you hit the fast mode limit, it falls back to standard speed and turns fast mode back on after the cooldown. Your own client won't do that unless you write it.

Moving off Sonnet 4.5

Sonnet 4.5 is the more likely surprise. It was a common default in agent frameworks, editor plugins, and internal tools written in late 2025, and pinned model IDs tend to stay put until they break. Anything that hardcodes claude-sonnet-4-5-20250929 needs a new ID.

Anthropic names claude-sonnet-5-5 as the replacement. Run your evals or a few representative agent tasks on it before you switch production traffic. Newer models can differ in output length and tool-use habits, which matters if you parse responses strictly. If your tool calls Claude through Amazon Bedrock, Google Cloud, or a third-party editor, check that platform's own retirement notice. Its dates may not match Anthropic's.

A short checklist

1. Search code, settings, and CI for claude-opus-4-8 with fast mode, and for claude-sonnet-4-5.
2. Switch fast mode workloads to Opus 5.5 and drop the beta header on those calls.
3. Switch Sonnet 4.5 calls to claude-sonnet-5-5 after a quick eval.
4. Make sure admin model allowlists include Opus 5.5 and Sonnet 5.5.
5. Check that retries don't loop on errors that won't go away.

For the rest of what changed in Claude Code this month, see our Claude Code 2.1.311 guide. For what agent usage costs in practice, see real AI coding costs.

Sources

• Claude Code changelog: 2.1.316 fast mode deprecation notice.
• Claude API docs: Fast mode: deprecation, speed parameter, beta header, rate limits, and usage.speed.
• Claude Code docs: Fast mode: defaults, settings, cache behavior, and admin controls.
• Claude API docs: Pricing: standard and fast mode rates.
• Claude API docs: Model deprecations: Sonnet 4.5 retirement and Opus 4.8 status.