I tried out Claude Opus 5.5, which became available on Amazon Bedrock

I tried out Claude Opus 5.5, which became available on Amazon Bedrock

These are the validation results from calling Claude Opus 5.5 via the Bedrock Converse API. I have summarized how to specify JP/Global profiles, the presence or absence of reasoningContent by effort level, the lack of Fast mode support, and the changes in pricing compared to Opus 5.
2026.09.23

This page has been translated by machine translation. View original

Introduction

On September 22, 2026, Anthropic announced Claude Opus 5.5, which also became available on Amazon Bedrock the same day.

https://www.anthropic.com/news/claude-opus-5-5

Here are the results of testing Opus 5.5 via Bedrock Runtime through an Inference Profile from the Tokyo region. The following information was verified on September 23, 2026, using AWS CLI 2.36.7.

Model Availability

I ran list-foundation-models in ap-northeast-1 and confirmed the Opus 5.5 entry.

{
  "modelId": "anthropic.claude-opus-5-5",
  "modelName": "Claude Opus 5.5",
  "providerName": "Anthropic",
  "inferenceTypesSupported": ["INFERENCE_PROFILE"],
  "modelLifecycle": {
    "status": "ACTIVE",
    "startOfLifeTime": "2026-09-22T10:00:00+00:00"
  }
}

The two Opus 5.5 profiles confirmed via list-inference-profiles from ap-northeast-1 are as follows.

Profile Name inferenceProfileId Routing Destination
JP Anthropic Claude Opus 5.5 jp.anthropic.claude-opus-5-5 ap-northeast-1, ap-northeast-3
Global Anthropic Claude Opus 5.5 global.anthropic.claude-opus-5-5 All supported regions

Calling via the Converse API

This is a minimal call specifying the JP profile. Aside from passing the profile ID to --model-id, the Converse API syntax is the same as for other Claude models.

aws bedrock-runtime converse \
  --region ap-northeast-1 \
  --model-id jp.anthropic.claude-opus-5-5 \
  --messages '[{"role":"user","content":[{"text":"自己紹介を1文で"}]}]'

The returned usage is as follows.

{
  "usage": {
    "inputTokens": 17,
    "outputTokens": 101,
    "totalTokens": 118,
    "cacheReadInputTokens": 0,
    "cacheWriteInputTokens": 0
  }
}

A response was also returned when replacing --model-id with global.anthropic.claude-opus-5-5. The usage when called with the Global profile is as follows.

{
  "usage": {
    "inputTokens": 17,
    "outputTokens": 97,
    "totalTokens": 114,
    "cacheReadInputTokens": 0,
    "cacheWriteInputTokens": 0
  }
}

Since the Global profile may route to regions anywhere in the world, please use the JP profile if you have requirements to keep data within Japan.

Behavior by effort Level

effort is not a standard parameter of the Converse API; it is passed as a model-specific field in additionalModelRequestFields. The valid values for output_config.effort (per Anthropic documentation) are five levels: low / medium / high / xhigh / max, with the default being medium.

https://docs.anthropic.com/en/api/messages

aws bedrock-runtime converse \
  --region ap-northeast-1 \
  --model-id jp.anthropic.claude-opus-5-5 \
  --messages '[{"role":"user","content":[{"text":"Pythonで素数判定する関数を書いて"}]}]' \
  --additional-model-request-fields '{"output_config":{"effort":"low"}}'

This time, I measured three levels with the same prompt. Each value is from a single measurement. Whether xhigh / max are accepted by Bedrock has not been confirmed.

effort reasoningContent outputTokens
low No block 1,281
medium Block present (text is empty) 1,312
high Block present (text is empty) 1,488

With medium and high, a block with an empty reasoningText.text and a signature was returned. If post-processing is written assuming the existence of a reasoningContent block, choosing low means that block will not be included, so branching on the parsing side becomes necessary.

Fast Mode Availability

Anthropic's Fast mode is specified with service_tier: "fast". On the other hand, the values accepted by Bedrock's service_tier are reserved / priority / default / flex. Passing fast actually results in a ValidationException. At least within the scope of this test, Fast mode does not appear to be available.

Note that the Bedrock Opus 5.5 model card also states that only the Standard service tier is supported.

https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-5-5.html

Pricing

These are the values listed for Asia Pacific (Tokyo) on the Bedrock pricing page. The JP profile falls under Geo / In-region, and all items are 10% higher than Global Cross-region.

Item Global Cross-region Geo / In-region (JP)
Input $4.00 $4.40
Output $20.00 $22.00
Batch N/A N/A

(All units are $/MTok)

The Global values are the same as the official Anthropic API pricing. Compared to Opus 5 within the same Geo, Input / Output has decreased by 20% from $5.50 / $27.50.

https://www.anthropic.com/pricing

https://aws.amazon.com/bedrock/pricing/

Summary

Claude Opus 5.5 became available on Bedrock on the same day as Anthropic's official announcement. Pricing is also 20% cheaper for Input / Output compared to Opus 5.

For Opus-series models available with the JP profile, since Opus 5 was not offered, Opus 4.8 (jp.anthropic.claude-opus-4-8) had been the latest, but now Opus 5.5 (jp.anthropic.claude-opus-5-5) has become available. The AWS documentation also explicitly states "Keeps data within Japan regions." For Japanese workloads with data residency requirements, this becomes a straightforward migration candidate.

If you have no requirement to keep data within Japan, try Opus 5.5 with the Global profile, which is 10% cheaper; if you do have such requirements, use the JP profile.

https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-5-5.html


Claudeならクラスメソッドにお任せください

クラスメソッドは、Anthropic社とリセラー契約を締結しています。各種製品ガイドから、業種別の活用法、フェーズごとのお悩み解決などサービス支援ページにまとめております。まずはご覧いただき、お気軽にご相談ください。

サービス詳細を見る

Share this article

AWSのお困り事はクラスメソッドへ