How to turn on fast mode for Claude Fable 5.2

Fast mode runs the same model with a faster inference configuration. There is no change to intelligence or capabilities. You can get up to 2.5x higher output tokens per second compared to standard speed. That gain is on output tokens per second, not time to first token, and it is most visible with streaming.

The opt-in is not the same on every surface. The Messages API takes a speed field. Managed Agents set that field on the model object. Microsoft Foundry uses a separate deployment and rejects speed on the non-fast model. Fast mode also has a rate limit of its own.

Set speed on the Messages API

Fast mode works only on supported models. On the Messages API, set speed to "fast".

The page’s Python sample uses these fields. Claude Fable 5.2 supports fast mode, but this page does not place a non-Foundry Fable model ID in a code sample. Keep the Fable 5.2 model ID your account already uses, and add speed. Do not invent an ID. The model string below is the one the page prints. Copy the surrounding sample from the fast mode page rather than reconstructing the client call.

    model="claude-opus-5-5",
    max_tokens=1024,
    speed="fast",
    messages=[{"role": "user", "content": "Hello"}],

A longer sample on the same page sets max_tokens to 4096 and uses the user text "Refactor this module to use dependency injection", then prints text blocks from the response. The shorter sample prints usage.speed. The comment on that line is "fast" or "standard".

When a request with speed: "fast" succeeds, usage.speed is "fast". The sample output includes this object:

{
  "id": "msg_01XFDUDYJgAACzvnptvVoYEL",
  "type": "message",
  "role": "assistant",
// ...
  "usage": {
    "input_tokens": 8,
    "output_tokens": 12,
    "speed": "fast"
  }
}

Requesting fast mode on a model that does not support it returns an error. Exceeding fast mode rate limits or capacity returns a 429 or a 529.

Claude Opus 4.6 is the documented exception, not Fable 5.2. With speed: "fast" and the fast-mode-2026-02-01 beta header, that request succeeds but runs at standard speed, and usage.speed shows "standard". On Claude Models on Google Cloud (Anthropic managed), the same Opus 4.6 request returns a 400 even with the header.

The data-handling note names the applicable endpoint as /v1/messages with speed: "fast". Zero data retention is eligible, excluding Covered Models, and applies regardless of the speed setting. HIPAA is listed as eligible.

Managed Agents

Set speed on the agent’s model object. The documented example is {"id": "claude-opus-5-5", "speed": "fast"}.

The availability table limits Managed Agents to Claude Opus 5.5, Claude Opus 5, and Claude Opus 4.8. It does not list Claude Fable 5.2 or Claude Mythos 5.2. A Fable Messages API opt-in is not a Managed Agent opt-in.

Foundry model IDs

On Claude in Microsoft Foundry, fast mode is a separate model. The page names claude-opus-5-5-fast, claude-fable-5-2-fast for Claude Fable 5.2, and claude-mythos-5-2-fast for Claude Mythos 5.2.

Send the request to a deployment of the fast model, and leave out speed. A request that sets speed on a deployment of the model without -fast returns an error.

The availability table splits Foundry by host. Microsoft Foundry hosted on Azure lists Claude Opus 5.5. Microsoft Foundry hosted on Anthropic lists Claude Fable 5.2 and Claude Mythos 5.2. Both notes use the same rule: the fast model ID ends in -fast, and you leave speed out. If your Fable deployment does not end in -fast, do not add the field.

The separate rate limit

Fast mode’s rate limit is separate from standard rate limits. Over the limit, the API returns a 429 with a retry-after header that says when capacity is available. On Claude Models on Google Cloud (Google managed), Google Cloud manages rate limits and quotas.

Claude Opus 5.5 has a default fast mode rate limit in the Start, Build, and Scale usage tiers, so you do not need to request access. Organizations in the Evaluation tier can use fast mode with Opus 5.5 once they move to Start, Build, or Scale. The same applies to Claude Fable 5.2. It also applies to Claude Mythos 5.2.

These response headers report fast mode status:

anthropic-fast-input-tokens-limit: maximum fast mode input tokens per minute
anthropic-fast-input-tokens-remaining: remaining fast mode input tokens
anthropic-fast-input-tokens-reset: when the fast mode input token limit resets
anthropic-fast-output-tokens-limit: maximum fast mode output tokens per minute
anthropic-fast-output-tokens-remaining: remaining fast mode output tokens
anthropic-fast-output-tokens-reset: when the fast mode output token limit resets

View them on the Rate limits page in the Claude Console. This page does not print the numeric caps.

By default, the SDK retries fast mode 429 errors up to two times. Change that with max_retries. Each retry waits for the retry-after delay. Fast mode uses continuous token replenishment, so that delay is typically short, and the request succeeds once capacity is available.

Pricing

Fast mode is priced at a multiplier on standard rates across the full context window, including requests over 200k input tokens. Rates published on this page:

Claude Fable 5.2: $16 USD / MTok input, $80 USD / MTok output
Claude Mythos 5.2: $16 USD / MTok input, $80 USD / MTok output
Claude Opus 5.5: $8 USD / MTok input, $40 USD / MTok output
Claude Opus 5 and Claude Opus 4.8: $10 USD / MTok input, $50 USD / MTok output

Prompt caching multipliers apply on top of fast mode pricing. Data residency multipliers apply on top as well. On Claude Models on Google Cloud (Google managed), Google Cloud sets fast mode prices. Those figures are not on this page.

For organization-wide usage and cost, the docs point to the Usage and Cost Analytics API.

Availability, and what to skip

On the Claude API, Messages API fast mode lists Claude Fable 5.2, Claude Mythos 5.2, and Claude Opus 5.5. Claude Opus 5 and Claude Opus 4.8 are research preview and require the fast-mode-2026-02-01 beta header. The page names that header and does not show how to attach it. For fast mode on Claude Opus 5, contact your account manager or join the waitlist. If you used fast mode on an earlier model, migrate to a model that still supports it. Migration steps are not on this page.

Claude Models on Google Cloud (Anthropic managed) lists the Messages API for Fable 5.2, Mythos 5.2, and Opus 5.5. The Google-managed listing includes Fable 5.2 and Mythos 5.2.

Fast mode is not available with the Batch API. It is not available with a Priority Tier commitment.

This page does not document a Claude Code toggle, or a dated fallback if a cloud host refuses fast mode. Those details are unknown here.

Fall back instead of waiting

To fall back to standard speed, catch the rate limit error and retry without speed: "fast". On Foundry, the deployment sets the speed, so retry on a deployment of the model without -fast.

Set max_retries to 0 on the initial fast request to skip automatic retries and fail immediately on a rate limit. That also disables retries for other transient errors, including overloaded and internal server errors, so the page’s Python example resends the request after those errors. The sample in that example uses model claude-opus-5-5 and max_tokens of 1024. On a Foundry Fable deployment, switch deployments instead of removing a speed field you never sent.

What to do next

Open the fast mode page and copy its Messages API sample, with speed set to "fast", then read usage.speed on the response. Check the six fast-mode headers, and open the Rate limits page in the Claude Console before you raise traffic. On Foundry, use claude-fable-5-2-fast only if Fable fast mode is listed for your hosting variant, and leave speed out. Re-read the pricing table before a long-context job: the premium covers the full window, including requests over 200k input tokens, and caching and data-residency multipliers stack on top.

Sources

Fast mode