Z.ai's GLM 5.3 model becomes available on Amazon Bedrock

On Amazon Bedrock, GLM 5.3 is accessible via fully managed APIs that support cross-Region inference, prompt caching, and various service tiers. The model supports both OpenAI-compatible APIs and native Bedrock endpoints. A key feature is prompt caching, which reduces latency and input costs for workloads that resend large context blocks, such as agentic workflows. AWS provides a walkthrough demonstrating how to use GLM 5.3 with Strix, an open-source AI penetration testing agent, to perform authorized security tests on applications like OWASP Juice Shop. This setup allows inference to run under the user's AWS account controls, avoiding third-party providers.

Access is currently limited to eligible enterprise customers. The model is available through US and Global cross-Region inference profiles. While the article highlights the model's strengths in long-horizon agentic tasks and complex systems engineering, it notes that users must have appropriate IAM permissions and, for the security demo, specific software like Docker and Strix installed. AWS emphasizes that unauthorized security testing is illegal and violates its Acceptable Use Policy.