On October 30, Anthropic deprecated fast mode for Claude Opus 4.8. It will be removed on November 30, 2026. After that, a request to claude-opus-4-8 with speed: "fast" returns an error from the API. To keep the speed, you migrate to Claude Opus 5.5.
The entry sits in the Claude Platform release notes and, word for word, in the Claude Code 2.1.316 changelog.
The behavior after the cutoff is the point
Deprecations are routine. This one is slightly less so, because of a single sentence: after removal, a fast mode request does not fall back to standard Opus 4.8 — it fails. If you have speed: "fast" sitting in a config and leave the model alone, December 1 does not bring you a slower run. It brings you a broken one.
That is the more expensive version of a silent failure, and the more honest one. A quiet fallback would have shifted bills and latencies with nobody noticing.
What fast mode buys and what it costs
Fast mode delivers up to 2.5x higher output tokens per second at premium pricing. Intelligence and capabilities do not change; it is the same model running a faster inference configuration. The gain is in output tokens per second, not time to first token, and it shows most clearly when you stream.
Fast mode is available on Claude Opus 5.5 and in research preview on Claude Opus 5 and Claude Opus 4.8 — so the research preview is exactly the piece going away. Opus 5.5 ships with a default fast mode rate limit in the Start, Build and Scale usage tiers, so there is no access request to file. That limit is separate from the standard Opus limit; exceeding it returns a 429 with a retry-after header.
Research preview means research preview
Nothing that ends here was ever promised. Fast mode on 4.8 ran as a preview, and previews get switched off once the successor carries the feature properly. The move still reads as a signal: Anthropic is clearing speed paths off the older Opus generations and consolidating them on 5.5. Anyone still running production on 4.8 gets a deadline until November 30 and a decision after it.