Loading page…
The paid setting promises faster output, not a smarter model. It costs $8 per million input tokens and $40 per million output tokens, with separate capacity limits.
Listen to this story
On October 15, Anthropic made Opus 5.5 fast mode available without preview approval, turning a speed setting into a paid option rather than a model upgrade: it uses the same weights and behavior. Developers can pay for generation Anthropic says reaches up to 2.5 times the standard output-token rate, but this does not reduce the wait for the first token. Access remains platform-specific, and separate fast-mode capacity limits can still throttle requests.
Fast-mode rates are $8 per million input tokens and $40 per million output tokens, including requests above 200,000 input tokens; caching and data-residency modifiers add charges.
Fast mode works through the Claude API and Claude Managed Agents, plus Microsoft Foundry for Azure-hosted deployments; it is unavailable on Amazon Bedrock and Google Cloud.
Fast-mode limits are separate from standard Opus limits; a 429 response includes a retry delay, and the SDK retries up to twice by default.
Developers no longer need to wait for permission to use Claude Opus 5.5 fast mode. Anthropic ended its research preview and removed the waitlist on October 15. The change opens access to a paid speed setting that, according to the company, produces output faster without changing the underlying model’s intelligence or capabilities.
Opus 5.5 fast mode now requires neither an access request nor the fast-mode-2026-02-01 beta header, an extra field used for preview access. Existing requests that send that header will continue working. Anthropic also assigns a default fast-mode rate limit at every usage tier.
response = client.messages.create(
model="claude-opus-5-5",
max_tokens=4096,
speed="fast",
messages=[{"role": "user", "content": "Refactor this module to use dependency injection"}],
)The removal of access restrictions applies specifically to Opus 5.5. Fast mode for Opus 5 and Opus 4.8 remains in research preview, requiring access through an account manager or waitlist, plus the beta header.
Anthropic says fast mode uses the same model weights and behavior, but a faster setup for generating answers. It claims up to 2.5 times as many output tokens per second. That gain concerns how quickly an answer is produced once output begins, not the wait before its first token.
Fast mode carries premium pricing across the full context window, including requests exceeding 200,000 input tokens. Prompt-caching and data-residency pricing modifiers apply on top of the listed rates in Anthropic’s fast-mode documentation.
Fast mode is available through the Claude API, including Claude Managed Agents. Opus 5.5 fast mode also works through Claude in Microsoft Foundry for Azure-hosted deployments, but not Anthropic-hosted deployments there. It is unavailable on Amazon Bedrock, Claude Platform on AWS and Google Cloud.
Fast-mode rate limits are separate from standard Opus limits. Exceeding them produces a 429 error with a retry-after header indicating when capacity will be available. The software development kit retries these requests up to twice by default, waiting for the server’s specified delay.
An application can instead catch the rate-limit error and retry without the fast setting. That fallback is developer-controlled, not automatic. Switching speeds also causes a prompt-cache miss: fast and standard requests do not share saved prompt prefixes. Successful fast requests report usage.speed as "fast" in the response.
Loading discussion...
Join the conversation
Explain when shorter waits would justify the added cost.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.