Anthropic Announces End of Research Preview for Claude Opus 5.5 Fast Mode and Cancels Waitlist
Anthropic announced the end of the research preview for Claude Opus 5.5 fast mode and the cancellation of the waitlist, with the same model capabilities unchanged and output speed improved by up to 2.5 times.
API developers are charged per token, at $8 per million input tokens and $40 per million output tokens, exactly double the standard speed of Opus 5.5; it can now be enabled via parameters without the need for an application.
Pro, Max, and other subscription users can enable it in Claude Code via /fast, and the costs do not count against their plan limits, being deducted directly from additional usage; team and enterprise versions require admin activation, and this rule remains unchanged.
The fast mode is now available on Microsoft Foundry (Azure hosted) and will soon be supported by Google Cloud; some parts of Claude Code still use additional usage credits.
The standard Opus 5.5 API price is $4 per million input tokens and $20 per million output tokens, with the fast mode billed at double rates across the entire context window.
In market mechanics, buyers are developers and subscription users needing faster interaction speeds, while the seller is Anthropic's inference capacity; the event is driven by the cancellation of the waitlist, with funds flowing from standard usage to the additional paid fast path, benefiting users willing to pay a premium for speed, while putting pressure on users relying on plan limits.
This model splits the same model into standard and fast payment paths, accelerating the transition from preview to publicly available.
Source: Public Information
ABAB AI Insight
Previously, Anthropic had positioned the fast mode of the Opus series as a research preview requiring application or waitlisting; this time, it has lifted restrictions on Opus 5.5 on the Claude Platform while maintaining the rule of charging from additional usage in Claude Code.
The capital path reflects the configuration of speed enhancements as a separate high-priced tier, with standard prices of $4/$20 and fast prices of $8/$40, allowing users willing to pay a premium to gain priority access to inference resources, while subscription users who do not enable additional usage cannot access it.
This is similar to the tiered pricing paths in cloud services for standard and accelerated instances, and resembles the practice of model providers charging for the same weights through different service configurations; it is currently transitioning from preview quota management to publicly tiered pricing.
Essentially, this represents a transfer of pricing power: Anthropic separates faster inference capacity from the unified plan through double rates, making speed a separately purchasable premium attribute, thus prioritizing resource allocation to additional paying users rather than distributing it evenly based on subscription limits.
ABAB News · Cognitive Laws
Same model split into two price lines
Speed premium does not count against plan limits
Cancelling the waitlist equals selling capacity to those willing to pay