Anthropic ends Claude Opus 4.8 fast mode on 30 November 2026

On 30 October 2026, Anthropic's Claude Platform release notes said fast mode for Claude Opus 4.8 is deprecated and will be removed on 30 November 2026. After that date, Anthropic says, requests to claude-opus-4-8 with speed: "fast" will return an error. The notes tell teams who want to keep fast mode to migrate to Claude Opus 5.5.

What breaks, and what does not

Fast mode is a request setting that runs the same model on a faster inference configuration. Anthropic's documentation says it gives up to 2.5 times higher output tokens per second at premium pricing, with no change to intelligence or capabilities. The speed gain is in output tokens per second, not time to first token, according to the same page.

The deprecation is about the fast setting on one model, not the model itself. The 30 October notes do not say Opus 4.8 is being retired, and I did not find a retirement date for it. Requests to claude-opus-4-8 without speed: "fast" are not mentioned in the notice.

The failure mode is the part to check. Anthropic's fast mode page says a request for fast mode on a model that does not support it returns an error. A team that hard-codes speed: "fast" for Opus 4.8 in a product feature or an agent loop will see failed requests from 30 November, not slower ones.

What the move to Opus 5.5 costs

Anthropic's pricing table on the fast mode page lists Claude Opus 5.5 at $8 per million input tokens and $40 per million output tokens. Claude Opus 5 and Claude Opus 4.8 are listed at $10 and $50. On those fast-mode rates alone, the 5.5 line is 20% cheaper per token for both input and output.

That comparison is of rate cards only. It says nothing about how many tokens Opus 5.5 uses on your tasks, and Anthropic's page does not give a like-for-like cost per task. The page also says fast mode pricing stacks with prompt caching and data residency multipliers.

Other conditions on the page

• Fast mode has its own rate limit, separate from standard Opus limits. Exceeding it returns a 429, and the SDK retries those automatically up to 2 times by default.
• Fast mode is not available with the Batch API or with a Priority Tier commitment.
• Requests at different speeds do not share cached prefixes, so falling back from fast to standard causes a prompt cache miss.
• Opus 5.5 has a default fast mode rate limit in the Start, Build and Scale usage tiers, so those organisations do not need to request access. Evaluation-tier organisations must move up a tier first.

What to check this week

Search your code and your agent configurations for speed: "fast" paired with claude-opus-4-8, including settings held in Managed Agents model objects. Test the same prompts on claude-opus-5-5 and compare output quality and cost per task, because the rate card does not settle that. Anthropic has a migration guide titled Migrating to Claude Opus 5.5 from Claude Opus 4.8 linked from the fast mode page.

What is not stated: whether Anthropic will email affected organisations beyond the documentation notice, or whether any grace period applies after 30 November. The notes give the date and nothing further.