Anthropic ends fast mode for Claude Opus 4.8 on Nov 30 and points to Opus 5.5

Anthropic has deprecated fast mode for Claude Opus 4.8 and will remove it on November 30, 2026, according to its Claude Platform release notes. After that date, a request to claude-opus-4-8 that asks for fast speed will fail with an error.

The entry is dated October 30, 2026. It tells developers who rely on the faster output setting to move to Claude Opus 5.5, which Anthropic launched on September 22.

What changes on November 30 for fast requests

Fast mode is a request option on the Messages API: setting speed: "fast" runs the same model on a faster inference configuration. Anthropic’s fast mode documentation says it delivers up to 2.5x higher output tokens per second, with no change to intelligence or capabilities. The gain applies to output speed, not time to first token, and is most visible when streaming.

The deprecation covers only the combination of Opus 4.8 and fast mode. The release notes do not say the model itself is being retired, and standard-speed requests to Opus 4.8 are not mentioned. What the page does state is that a fast request on that model will fail once the removal date passes, rather than quietly running at standard speed. Anthropic’s documentation notes that other cases behave differently: for Claude Opus 4.6 on the Claude API, a fast request succeeds but runs at standard speed.

Anyone who has hard-coded speed: "fast" against Opus 4.8 therefore has until the end of November to change the model name or drop the parameter. The documentation says the response’s usage.speed field reports whether a request ran at "fast" or "standard" speed, which gives teams a way to check what they are getting today.

Opus 5.5 is cheaper per token in fast mode

The replacement is not just a newer model, it is also less expensive in fast mode. According to Anthropic’s fast mode page, fast pricing is $8 per million input tokens and $40 per million output tokens on Claude Opus 5.5, against $10 and $50 for Claude Opus 5 and Claude Opus 4.8. Standard Opus 5.5 pricing is $4 and $20 per million tokens, per the release notes, so the fast premium works out to twice the standard rate. The premium applies across the full context window, and caching and data residency multipliers stack on top.

Access has also changed. On October 15, Anthropic moved fast mode for Opus 5.5 out of research preview. Requests to that model no longer need the fast-mode-2026-02-01 beta header, although sending it does no harm. Opus 5.5 also gained a standard fast mode rate limit for accounts in the Start, Build and Scale usage tiers. Opus 5 and Opus 4.8 remain in research preview, so requests to those two models should keep sending the header. Fast mode for Opus 5.5 is available on the Claude API, on Claude Models on Google Cloud (Anthropic managed) and on Claude in Microsoft Foundry for deployments hosted on Azure, according to the release notes and the fast mode page.

Who has to act, and what the swap involves

The angle that matters most is practical: this is a migration job for developers, and Opus 5.5 is not a drop-in rename. The release notes for its launch say that thinking cannot be switched off on Opus 5.5, so both thinking: {"type": "disabled"} and thinking: {"type": "enabled", ...} produce a 400 error. Forcing a tool call through the any or tool settings of tool_choice is rejected the same way. Computer use on the Claude API and Google Cloud now needs the computer_toolset_20260801 toolset. Code written for Opus 4.8 that forces tool calls or sets a thinking budget will need edits before the fast mode switch works. Anthropic points to a migration guide for Opus 5.5 from Opus 4.8.

The fast mode page lists further limits that affect planning. Fast mode is not available with the Batch API or with a Priority Tier commitment, and it has its own rate limit, separate from standard Opus limits. When that limit is exceeded, the API returns a 429 error with a retry-after header. Teams that fall back to standard speed on a 429 should expect a prompt cache miss, because requests at different speeds do not share cached prefixes.

The deprecation also fits a pattern in the same release notes. Anthropic announced on September 30 that it will retire Claude Sonnet 4.5 on the Claude API on November 30, 2026, the same date. Developers on older Claude models face a crowded end of November.

What the source does not establish

The release notes give no reason for the deprecation. They do not say how many customers still use fast mode on Opus 4.8, and they do not explain why the removal comes only weeks after Opus 5.5 left research preview. The 2.5x figure is Anthropic’s own and is stated as an upper bound; the page offers no independent measurement, and real gains will depend on the workload.

The same notes show Anthropic’s API changing quickly in other places too. Earlier this week it raised the Claude API request limit to 128 MB for Messages and Token Counting. For fast mode users, the immediate task is narrower: pick a date before November 30, test on Opus 5.5, and update the model name.