AWS has announced the availability of Claude Opus 5, Anthropic’s most advanced Opus-tier model, on Amazon Bedrock and the newly integrated Claude Platform on AWS.
The launch marks the first release in Anthropic’s fifth-generation model lineup, bringing substantial upgrades to agentic coding, long-running autonomous tasks, and cybersecurity-relevant reasoning at what AWS describes as Opus-tier pricing with Fable 5-level intelligence.
For security teams evaluating AI infrastructure, the most notable detail is Bedrock’s zero data retention (ZDR) by default for Claude Opus 5.
AWS Launches Claude Opus 5 With Advanced Cyber Capabilities
This removes a common blocker for regulated industries deploying LLMs in production, ensuring prompts and outputs aren’t persisted beyond the inference call.
The model runs on Bedrock’s next-generation inference engine, allowing organizations to maintain regional data residency and enterprise governance controls while scaling inference with zero operator access to underlying data.
The parallel Claude Platform on AWS offers ZDR on request and mirrors Anthropic’s native API experience directly inside the AWS Console, unified under existing AWS billing and IAM-based authentication.
This is a relevant detail for organizations managing compliance across multi-cloud AI deployments, where consistent authentication and billing pathways simplify audit trails.
According to AWS, Opus 5 delivers measurable gains over Claude Opus 4.8 in both coding and cybersecurity-specific tasks. The model can now navigate large codebases autonomously, adapting its strategy mid-task the way an experienced engineer would when encountering unexpected obstacles.
It also powers reliable long-running agents capable of working for hours or overnight, recovering from errors and finding alternative paths toward a goal rather than stalling out.
On the reasoning side, Opus 5 shows improved comprehension of long documents, with the largest accuracy gains appearing on document-heavy enterprise workloads such as compliance review and structured analysis.
AWS also highlights better sub-agent task decomposition, meaning complex jobs can be broken into smaller pieces that require less human oversight to complete correctly.
Notably, AWS states that in higher-risk scenarios, Opus 5 can automatically fall back to Claude Opus 4.8, with users notified when this occurs and configurable override options available through the API.
This fallback behavior is likely tied to Anthropic’s responsible scaling policies, which restrict the model from generating certain offensive-capable outputs without additional safeguards in place.
Developers can access Claude Opus 5 through the Bedrock console Playground or programmatically through the Anthropic Messages API via bedrock-runtime or bedrock-mantle endpoints, as well as through the standard Invoke and Converse APIs using the AWS CLI and SDKs.
A notable new feature is support for mid-conversation tool management, allowing developers to add or remove tools dynamically using tool_addition and tool_removal content blocks on system-role messages instead of resending the full tools array on every turn.
This is particularly useful for agentic security tooling that provisions capabilities like vulnerability scanners only when a given task requires them.
Claude Opus 5 is live on Amazon Bedrock in US East (N. Virginia), Asia Pacific (Melbourne), Europe (Ireland), and Europe (Stockholm), with broader coverage on Claude Platform on AWS spanning North America, South America, Europe, and Asia Pacific.
AWS is also promoting its Advanced Prompt Optimization tool to help teams benchmark and rewrite prompts for production use with the new model.
Cut SOC investigation blind spots and contain threats earlier to reduce response costs and business disruption with ANY.RUN.
