AWS has added Z.ai’s 753-billion-parameter GLM 5.3 to Bedrock, offering eligible enterprises managed access for coding and agentic workflows.
AWS announced on October 5, 2026, that Z.ai’s GLM 5.3 is available through Amazon Bedrock to eligible enterprise customers. The model has 753 billion parameters in a mixture-of-experts design and is intended for coding and extended agent workflows. Customers can use AWS-managed APIs rather than operating their own inference infrastructure.
Developers can access GLM 5.3 through OpenAI-compatible Responses and Chat Completions APIs, as well as Bedrock’s Invoke and Converse APIs. AWS recommends the OpenAI-compatible options for new applications because they support a broader set of features. The service also offers US and global cross-Region inference profiles, with Flex, Standard and Priority service tiers for different cost and latency needs.
Prompt caching is available to reduce repeated processing of stable context, such as repository files or long system instructions. Implicit caching is enabled by default; developers can also mark reusable prompt sections explicitly. AWS says an explicitly marked prefix must contain at least 1,024 tokens to qualify for caching.
Performance figures in the announcement are attributed to Z.ai. The company reports a 50% improvement over GLM 5.2 on its internal coding benchmark and a score of 84.5 on CyberGym at release. AWS notes that Z.ai did not publish a direct comparison with GLM 5, saying benchmark tests had been updated since the GLM 5.1 announcement. These are company-reported results, not independent AWS evaluations.
AWS also demonstrates using GLM 5.3 with Strix, an open-source penetration-testing agent, against a deliberately vulnerable application running locally. The example is framed as authorized testing: AWS says users should test only systems they own or have written permission to assess. The Bedrock offering combines model access with managed inference, caching and API options, while leaving application design and security-testing authorization with customers.