Claude Opus 5.5 fast mode leaves research preview: no waitlist, $8 / $40, now on Microsoft Foundry too

Anthropic took fast mode for Claude Opus 5.5 out of research preview on October 15 and removed the waitlist. Any Claude API user can now set speed: "fast" on an Opus 5.5 request with no access request and no beta header; requests that still send the old header keep working. Anthropic's developer account announced it on X at 20:09 UTC, and the fast mode docs say the same.

THE PRICE

$8 per million input tokens and $40 per million output, double Opus 5.5 at standard speed ($4 / $20), across the whole context window, including prompts over 200K tokens. Prompt caching and data residency multipliers apply on top, so a cache hit costs 5% of the fast input price, $0.40. The price has not changed since the research preview.

HOW FAST

Up to 2.5x the output tokens per second of Opus 5.5 at standard speed, by Anthropic's own figure. It is the same model: only the serving speed and the price change. Anthropic's docs say the gain is in output speed, not time to first token.

WHERE IT RUNS

On the Claude API, including Claude Managed Agents, and on Claude in Microsoft Foundry for deployments hosted on Azure. Not on Claude Platform on AWS, Amazon Bedrock or Google Cloud yet; Anthropic's post says Google Cloud is coming soon. It does not work with the Batch API or a Priority Tier commitment, and it has its own rate limit, separate from standard Opus, with a default in the Start, Build and Scale usage tiers. Switching between fast and standard speed throws away the prompt cache.

Fast mode in Claude Code is unchanged and still draws on extra usage credits, Anthropic says.

WHAT DID NOT CHANGE

Fast mode on Claude Opus 5 and Opus 4.8, at $10 / $50, is still a research preview behind the waitlist and the beta header.

AGAINST OPENAI

GPT-6.1 Sol Ultrafast, open to every OpenAI API user since October 8, was $12 / $60 per million tokens for prompts up to 272K when we compared the two on October 10. Opus 5.5 fast mode is a third cheaper on input and output, and the access gap we flagged then is gone.

WHAT WE DO NOT KNOW

How fast either one runs in practice: both speed claims are the vendors' own, and we have not seen an independent measurement. Also how large the default fast mode rate limits are for each tier, and when Google Cloud gets it.