Anthropic Claude Haiku 5.5 is now available on Amazon Bedrock and Claude Platform on AWS, positioned as the fastest model in the Claude 5.5 family for subagents and high-volume work. For most tasks it costs around 75 percent less than Claude Haiku 4.5, according to Anthropic. The release matters for business because it lowers the price of running agents that make many small calls per day.
Two access paths and effort controls
Amazon offers two routes to the model. Amazon Bedrock keeps data within AWS infrastructure with Regional data residency and connects to IAM for access, CloudTrail for audit, CloudWatch for monitoring, and Amazon Bedrock Guardrails, with usage shown on the AWS bill. Claude Platform on AWS provides direct access to the native Anthropic platform experience through the AWS Management Console, with the same APIs, features and console experience as working with Anthropic directly, combined with AWS billing and authentication.
Haiku 5.5 is described as the most capable Haiku model across coding, tool use, computer use and agentic tasks. It is the first Haiku model with effort controls, which let teams tune cost against intelligence for each task rather than using one setting for a whole workload. The model handles agentic coding and multi-step tool use and supports high-resolution images. That combination points to delegation of narrow steps inside larger workflows instead of assigning every step to a large model.
The timing links to the pairing with Claude Opus 5.5, announced recently. In that setup Opus 5.5 plans and makes judgment calls, while Haiku 5.5 carries out well-defined tasks quickly and at scale. Availability runs through US, EU, AU, JP and Global inference profiles on bedrock-runtime, plus bedrock-runtime and bedrock-mantle endpoints in AWS GovCloud (US). Claude Platform on AWS availability is listed for North America, with pricing detailed in Amazon Bedrock pricing.
What this means for high-volume AI work
For companies running support assistants, document processing or coding helpers, the practical shift is cost per action. Haiku 5.5 covers routing requests, code review, classification of long documents, extraction from small-to-medium documents, initial scans and quick answers over a knowledge base. Small firms can add such features to production without building separate infrastructure, while large firms can split workloads so expensive reasoning runs only where judgment is needed.
The limits sit in task selection and verification. Fast response fits simple conversations and repetitive browser and desktop steps, plus quick UI and UX iterations and small specific changes across multiple files. It does not replace careful review for complex decisions, ambiguous documents or large refactors. Before rollout, teams should test accuracy on their own data, check guardrails behavior, and track usage, performance and costs in CloudWatch and Cost Explorer as demand grows.
The marker to watch is how quickly teams move to a two-model pattern with Opus 5.5 for planning and Haiku 5.5 for execution in Bedrock logs and bills. Broader use of Playground tests, Boto3 InvokeModel calls and Converse API traffic would signal that the cheaper tier handles production volume. If that split holds, agent budgets will stretch further without added latency.
