What changed
ClaudeDevs calls it the same Opus 5.5, up to 2.5x faster, at $8 / $40 per million tokens. The fast-mode page states that as up to 2.5x higher output tokens per second than standard speed, with time to first token outside the figure. The page says fast mode runs the same weights and behavior under a faster inference configuration, with no change to intelligence or capabilities, and that streaming is where the gain shows.
The post's $8 / $40 pair is input, then output. The page says fast-mode pricing is a multiplier on standard rates across the full context window, including requests over 200k input tokens. Prompt-caching and data-residency multipliers apply on top. The page does not print those extra percentages.
| Model | Input | Output | Access |
|---|---|---|---|
| Claude Opus 5.5 (claude-opus-5-5) | $8 USD / MTok | $40 USD / MTok | Open on the Claude API, including Claude Managed Agents. No beta header or access request. |
| Claude Opus 5 (claude-opus-5) and Claude Opus 4.8 (claude-opus-4-8) | $10 USD / MTok | $50 USD / MTok | Research preview. Account manager or waitlist, plus the fast-mode-2026-02-01 beta header. |
Who can use it
Open access is claude-opus-5-5 on the Claude API, including Claude Managed Agents. The page says Opus 5.5 has a default fast-mode rate limit in every usage tier. claude-opus-5 and claude-opus-4-8 stay in research preview. Callers need an account manager or the waitlist, plus the fast-mode-2026-02-01 header.
The page says fast mode is unavailable on Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. ClaudeDevs says Opus 5.5 fast mode is already available in Microsoft Foundry, hosted on Azure, and that Google Cloud is coming soon. The page gives Google Cloud no date. ClaudeDevs also says fast mode in Claude Code remains unchanged and uses extra usage credits. The fast-mode page does not describe Claude Code billing.
The Batch API and a Priority Tier commitment sit outside fast mode. On claude-opus-4-7, speed set to fast returns an error and does not fall back to standard speed. The model remains available at standard speed. On claude-opus-4-6, speed set to fast without the beta header returns a 400. With the header, the request runs at standard speed, bills at standard rates, and reports usage.speed as standard.
How to turn it on
Set speed to fast on claude-opus-5-5. Opus 5.5 needs no beta header. A request that still sends fast-mode-2026-02-01 keeps working. A successful fast response reports usage.speed as fast. Research-preview models still need that header.
Fast mode has its own limits, separate from standard Opus limits. Past the limit or the capacity cap, the API returns a 429 or a 529. A rate-limit response is a 429 with a retry-after header. The SDK retries that 429 up to two times by default (max_retries). The response includes anthropic-fast-input-tokens-limit, anthropic-fast-input-tokens-remaining, anthropic-fast-input-tokens-reset, anthropic-fast-output-tokens-limit, anthropic-fast-output-tokens-remaining, and anthropic-fast-output-tokens-reset. Numeric caps for an account are on the Rate limits page in the Claude Console.
Switching a retry from fast to standard speed misses the prompt cache. The two speeds do not share cached prefixes.
What is still open
The fast-mode page states the Opus 5.5 access rule and does not date a waitlist removal. The 15 October timestamp is when ClaudeDevs published the post. Per-tier tokens per minute are not printed on the fast-mode page. Zero data retention is not stated there either. The page points that question to a separate API and data retention document.
Source
Also from
Sources and editorial approach
Sources for this brief are the ClaudeDevs post published at 20:09 UTC on 15 October 2026 (https://x.com/ClaudeDevs/status/2110825294655562048) and the Claude Platform fast-mode page (https://platform.claude.com/docs/en/build-with-claude/fast-mode), both read on 15 October 2026. The Claude Code changelog (https://code.claude.com/docs/en/changelog.md) was read from the top through the recent release notes for any statement that Claude Code fast mode is unchanged or billed as extra usage. Those notes do not state that rule, and the final portion of the changelog page was not included. The Pricing page, the Console rate-limits page, the API and data retention page, and an earlier ClaudeDevs post linked from the announcement were not covered. No speed measurement, invoice check, or cloud-console check was run.