AWS announced on September 22 that Claude Opus 5.5 is available through Amazon Bedrock and Claude Platform on AWS. It is the first model in Anthropic’s Claude 5.5 family and is positioned by AWS for agentic coding, knowledge work and long-running tasks.
According to Anthropic’s claims cited by AWS, Opus 5.5 completes work with fewer tokens than Opus 5. AWS says lower per-token prices and cheaper cache reads further reduce the average cost per task. The announcement does not state specific rates, instead directing customers to Amazon Bedrock’s pricing information.
Adaptive thinking is permanently enabled, allowing the model to decide how much reasoning a task requires. Developers can use the effort control rather than setting manual thinking budgets. AWS also says the model communicates more clearly about its actions, findings and outstanding needs, which is intended to make longer tasks easier to follow.
Opus 5.5 is the first Opus model to include safety classifiers similar to those used by Claude Fable 5.1 for biology, cybersecurity and AI development. AWS explicitly cautions that the new model will refuse requests more frequently than earlier Opus versions, a limitation teams should consider when evaluating existing workflows.
Customers can test the model in the Amazon Bedrock Playground or call it programmatically. Supported options include the Anthropic Messages API through the Anthropic SDK, plus the Invoke and Converse APIs through AWS SDKs or the AWS CLI. AWS examples use the model identifier global.anthropic.claude-opus-5-5. Because a response can place a thinking block before its text, AWS advises selecting the block by type rather than by a fixed array position.
On bedrock-runtime, access is provided through US, EU, AU, JP and Global inference profiles. The model also runs on bedrock-mantle in US East (N. Virginia) and AP Southeast (Melbourne). Claude Platform on AWS offers it in North America, while AWS refers readers to the Amazon Bedrock documentation for the complete regional list.
Practical context: For teams already using Amazon Bedrock, the release provides a way to assess Opus 5.5 through their existing AWS model-access options. A practical comparison with Opus 5 should account for full-task cost, safety-related refusal rates and the behavior of long-running agentic workflows. AWS presents vendor performance and efficiency claims but does not publish independent tests or detailed comparative results in this announcement.
| Environment | Availability | Interface |
|---|---|---|
| Amazon Bedrock | US, EU, AU, JP and Global profiles | bedrock-runtime |
| Amazon Bedrock | US East (N. Virginia), AP Southeast (Melbourne) | bedrock-mantle |
| Claude Platform on AWS | North America | Claude Platform |
Requirements for programmatic access
The AWS example assumes an active account with Amazon Bedrock access, Python 3.10 or later, a configured AWS CLI, the Boto3 and Anthropic packages, the Amazon Bedrock Token Generator, and the IAM permissions bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream.
Sources
Event date: 2026-09-22. Primary source date: 2026-09-22.
