Claude Opus 5.5 Fast Mode: Enable It With Request Examples

To use fast mode on Claude Opus 5.5, set speed: "fast" in a request to claude-opus-5-5. That's it. As of October 16, 2026, fast mode is out of research preview for this model, so you no longer need to request access, and you don't need the fast-mode-2026-02-01 beta header.

If your code already sends that header, leave it. Anthropic's release notes say requests that still send it keep working. The catch is that other models haven't graduated yet, which matters if you switch models in the same code path. Details are below.

The smallest request that turns on fast mode#

The request body is a normal Messages API body with one extra field, speed:

{
  "model": "claude-opus-5-5",
  "max_tokens": 1024,
  "speed": "fast",
  "messages": [
    { "role": "user", "content": "Summarize this stack trace in two sentences." }
  ]
}

Three things to check in that body:

• model is claude-opus-5-5, the Claude API ID listed on the models overview.
• speed is "fast". Leave the field out and you get standard speed.
• There's no thinking field. More on that in the gotchas section.

Nothing else changes. Same endpoint, same message format, same response shape.

What changed on October 15, 2026#

The release notes for October 15, 2026 bundle several changes for Opus 5.5 fast mode:

• Fast mode is out of research preview for Claude Opus 5.5. You no longer need to request access.
• Requests to Opus 5.5 no longer require the fast-mode-2026-02-01 beta header. Requests that still send it keep working.
• Opus 5.5 now has a default fast mode rate limit in the Start, Build, and Scale usage tiers.
• Fast mode for Opus 5.5 is available on Claude in Microsoft Foundry for deployments hosted on Azure.

Before this, fast mode on Opus 5.5 was a research preview that launched on September 22, 2026 alongside the model. Earlier fast mode models also needed a waitlist. Opus 5.5 no longer does.

Do you still need the beta header?#

Not for Opus 5.5. For two other models, yes. The release notes say fast mode for Claude Opus 5 and Claude Opus 4.8 remains in research preview, and to keep sending the beta header with fast mode requests to those models.

If your app lets users pick between these models, don't delete the header globally. Add it conditionally. This small script shows one way to build the request, and you can run it as is:

const NEEDS_FAST_BETA = new Set(["claude-opus-5", "claude-opus-4-8"]);

function fastRequest(model, prompt) {
  const body = {
    model,
    max_tokens: 1024,
    speed: "fast",
    messages: [{ role: "user", content: prompt }],
  };
  const betas = NEEDS_FAST_BETA.has(model) ? ["fast-mode-2026-02-01"] : [];
  return { body, betas };
}

for (const m of ["claude-opus-5-5", "claude-opus-5", "claude-opus-4-8"]) {
  const { body, betas } = fastRequest(m, "ping");
  console.log(m, "speed:", body.speed, "betas:", JSON.stringify(betas));
}

The logic is the point. Opus 5.5 gets no beta entry, and the two preview models keep theirs. When Opus 5 or Opus 4.8 graduate, you shrink the set. I'd keep the list in one place so that edit is a one-liner.

Check that the request actually ran fast#

The release notes describe a usage.speed field in the response that reports the speed used. They mention it in the context of Claude Opus 4.6, where fast requests silently ran at standard speed after fast mode was removed. The lesson transfers: don't assume. Log usage.speed next to your latency numbers, so a request that ran at standard speed shows up in your data instead of surprising you on the bill or in the p95.

Look at a real response from your own account before you build an alert on this field.

Fast mode pricing for Opus 5.5#

Fast mode costs more per token. The pricing page lists these rates as of October 16, 2026:

A few details from the same page:

• Fast mode pricing applies across the full context window, including requests over 200k input tokens.
• Prompt caching multipliers apply on top of fast mode pricing.
• Data residency multipliers apply on top of fast mode pricing.
• Fast mode is not available with the Batch API.

So the batch discount isn't an option for fast requests. If a job is offline and can wait, send it to batch at standard speed. Save fast mode for interactive paths where a person is watching tokens stream in.

The pricing page doesn't give a speed multiple for Opus 5.5 fast mode. The only number in the release notes is "up to 2.5x as fast," from the original Opus 4.6 launch. Measure your own workload before you promise anyone a speedup.

Where fast mode is available#

Per the pricing page, fast mode is available on the Claude API (first-party) and, for Claude Opus 5.5, on Claude in Microsoft Foundry for deployments hosted on Azure. It is not available on Claude Platform on AWS or partner-operated cloud platforms.

That last line trips people up. The model itself is listed on Amazon Bedrock, Google Cloud and other platforms on the models overview, but the speed: "fast" option isn't. If you route Opus 5.5 through one of those, plan for standard speed.

Rate limits#

Opus 5.5 now has a default fast mode rate limit in the Start, Build, and Scale usage tiers. The release notes don't give the numbers. Check your limits in the Claude Console rather than guessing. If you hit a limit, a sensible fallback is to retry that request without speed, but that's a design choice on your side, not documented behavior.

Common mistakes and gotchas#

Deleting the beta header everywhere. It's safe to drop for claude-opus-5-5 only. Opus 5 and Opus 4.8 fast mode still expect it.

Sending speed: "fast" to old models. Fast mode has been removed on older models, with different failure modes:

• claude-opus-4-7: since July 24, 2026, speed: "fast" returns an error. There's no fallback to standard speed.
• claude-opus-4-6: since June 29, 2026, the request runs at standard speed, is billed at standard rates, and doesn't return an error. That's why checking usage.speed matters.

Carrying old Opus request settings over. Opus 5.5 has behavior changes unrelated to fast mode that will still break a copied request:

• thinking: {"type": "disabled"} and thinking: {"type": "enabled", ...} return a 400 error. Omit thinking and use the effort parameter to control depth.
• tool_choice types any and tool return a 400 error. Use auto with strict tool use.
• On the Claude API and Google Cloud, computer use requires the computer_toolset_20260801 toolset. The earlier computer_20251124 tool returns a 400.

If a fast request fails with a 400, check these before you blame speed.

Expecting batch to work. Fast mode and the Batch API don't combine.

Assuming a waitlist. Older posts and older release notes say "join the waitlist." For Opus 5.5 that's out of date. The newest release notes say you no longer need to request access.

Expecting the same effort default. The models overview lists Opus 5.5's default effort as medium. If you compare fast and standard output quality, set effort explicitly on both so you're testing one variable.

Should you turn it on?#

My advice, not a documented rule: use fast mode where latency is felt by a user, such as chat, inline code edits, and interactive agents. Keep standard speed for background jobs, evals and anything you can batch. At $8 / $40 versus $4 / $20 per MTok, fast mode doubles the listed per-token rate, so a blanket switch roughly doubles your Opus 5.5 spend for the same traffic.

A cheap rollout path is a per-request flag. Add speed: "fast" for a slice of traffic, log usage.speed and latency, then compare cost against what the faster response is worth to your users.

For current model IDs, tiers and prices, the release notes, models overview and pricing page are the sources to check, since fast mode status has changed model by model over the past months.

Quick checklist#

• Model is claude-opus-5-5.
• Body includes "speed": "fast".
• No fast-mode-2026-02-01 header needed; harmless if present.
• Keep the header for claude-opus-5 and claude-opus-4-8.
• No thinking field, no tool_choice of any or tool.
• Not using the Batch API, Claude Platform on AWS or a partner cloud for these requests.
• Logging usage.speed and latency per request.
• Budget at $8 / $40 per MTok for fast requests.
• Rate limit checked in the Claude Console for your usage tier.

Quick answers