Claude Opus 5.5 fast mode no longer requires an access request
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
Anthropic now lets organizations in its API's Start, Build and Scale tiers run Claude Opus 5.5 in fast mode with no prior request. The model is the same, but it writes each answer faster, up to 2.5 times, and a million tokens now costs $8 on input and $40 on output. For Claude Opus 5 and Claude Opus 4.8 nothing changes, and there is still a waitlist.
On Thursday, October 15, Anthropic took the fast mode of Claude Opus 5.5 out of research preview on the Claude API, the route developers use to build its models into their own software. Since then there is no need to request access or add the beta header to requests, although those that still include it will keep working.
Organizations on the Start, Build and Scale usage tiers now get a default limit for this mode, separate from the one they have for Opus at normal speed. Those on the Evaluation tier will be able to use it once they move up to one of those three. If the limit is exceeded, the API replies with a 429 error stating when to try again, and each organization can check its quota in the Claude Console.
According to Anthropic's documentation, it is the same model running on a faster inference setup, with no change to its intelligence or to what it can do. It multiplies the output tokens (the chunks of text the model works with) produced each second by up to 2.5, an improvement that shows in how quickly it writes the answer, not in how long it takes to start.
In exchange, it costs twice as much as standard mode: $8 per million input tokens and $40 per million output tokens, compared with the $4 and $20 listed on its pricing page. Caching and data residency multipliers are added on top of that rate, and fast mode cannot be used with the Batch API.
Beyond the Claude API, it is available on Claude in Microsoft Foundry, Microsoft's service for using Claude, but only for deployments hosted on Azure: not for those hosted by Anthropic, nor on Claude Platform on AWS or other partner clouds. For Claude Opus 4.8 and Claude Opus 5, fast mode remains in research preview: access has to be requested from an account manager or through a waitlist, the beta header is still needed, and the price is $10 per million input tokens and $50 per million output tokens.
Why it matters · analysis and opinion
Going from a waitlist to a default rate limit turns fast mode into something a product can be planned around, not just tested. It makes sense wherever someone is watching the screen while the text appears: coding assistants, customer service chats or agents that draft long documents. Since the gain is in writing speed and not in how long the first word takes to arrive, short requests will barely notice it. Because it has its own quota, a usage spike can leave an application without fast mode even if it still has room at standard speed, and falling back to standard speed means rebuilding the cache. Before turning it on for every user, it is worth measuring with real traffic whether the speed makes up for the extra cost.
Official source: Anthropic · Written with the help of AI: how we make the news
