Anthropic Opens Claude Opus 5.5 Fast Mode to Everyone: Up to 2.5x Output Speed at $8/$40 per Million Tokens

Anthropic has taken fast mode for Claude Opus 5.5 out of research preview, removing the waitlist that gated access since the option debuted. Any Claude API customer can now request faster serving by setting a speed parameter on their requests, with no access application and no beta header required. Older clients that still send the retired beta header will keep working.

The pricing is a straightforward premium for speed: $8 per million input tokens and $40 per million output tokens, exactly double the standard Opus 5.5 rate of $4/$20. Prompt caching still stacks on top, so a cache hit against the fast input price costs $0.40. Anthropic says the same model weights and intelligence are served either way; only the inference configuration changes.

The headline claim is up to 2.5 times higher output tokens per second than standard Opus 5.5. That figure is Anthropic's own measurement and has not been independently verified. The company also specifies that the gain applies to output streaming speed, not time to first token, which matters for teams tuning interactive agents where the pause before a response starts is often what users feel most.

Availability is narrower than the model itself. Fast mode runs on the Claude API, including Claude Managed Agents, and on Claude in Microsoft Foundry for Azure-hosted deployments. It is not yet on Amazon Bedrock or Claude Platform on AWS, and Anthropic says Google Cloud support is coming soon. The option is excluded from the Batch API and Priority Tier commitments, carries its own rate limits separate from standard Opus, and switching between fast and standard speeds invalidates the prompt cache, a hidden cost worth pricing in for hybrid workloads.

The competitive context is a speed-pricing race. OpenAI opened GPT-6.1 Sol Ultrafast to every API user on October 8 at $12/$60 for prompts up to 272K tokens. Anthropic's fast mode undercuts that by roughly a third on both input and output, and is even cheaper per token than the fast modes of its own older models: Opus 5 and Opus 4.8 remain in research preview at $10/$50, still behind a waitlist, while fast mode for Opus 4.7 was retired in June.

For engineering teams running long agentic loops, the real test is tail latency rather than averages, and no third-party benchmarks of either vendor's speed claims have surfaced yet. But the structural signal is clear: serving speed has become a priced, marketing-worthy dimension of frontier API plans, alongside context length and price per token, and both major labs are now charging for it.