AWS adds GLM 5.3 to Amazon Bedrock for eligible enterprises

AWS has made Z.ai’s GLM 5.3 available through Amazon Bedrock for eligible enterprise customers. Eligible customers can run the model through OpenAI-compatible or native Bedrock APIs without operating their own inference infrastructure.

The announcement does not define AWS’s eligibility criteria. Customers with access can use the OpenAI-compatible Responses and Chat Completions APIs, or Amazon Bedrock’s Invoke and Converse APIs. For no-code access, users can select Test, then Playground, in the Bedrock console and choose GLM 5.3 from the model list. The launch follows AWS’s addition of Grok 4.7 to Bedrock.

AWS describes GLM 5.3 as a 753-billion-parameter mixture-of-experts model optimized for coding and long-horizon agentic tasks. Z.ai’s hosted model card reports the benchmark results discussed below. Amazon Bedrock exposes it through the US cross-Region profile us.zai.glm-5.3 and the Global profile global.zai.glm-5.3. AWS’s cross-Region documentation says geographic profiles keep processing within their geography, while global profiles can route requests to supported commercial AWS Regions worldwide. AWS says traffic between Regions stays on its network and is encrypted in transit.

The Bedrock integration supports implicit prompt caching by default and explicit cache controls through the Responses and Chat Completions APIs. Prompt caching reuses repeated prompt material, such as system instructions or repository context, across calls. AWS says this can reduce input costs and response latency.

Customers can choose Flex for lower-cost, less time-sensitive work; Standard for the default balance of price and speed; or Priority for latency-sensitive requests at a higher price. AWS lists three IAM permissions for its examples: bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream and bedrock:CallWithBearerToken.

The announcement does not provide a GLM 5.3 price table or enumerate every source and destination Region supported by the two inference profiles.

Performance figures remain vendor claims. Z.ai says GLM 5.3 improved 50% over GLM 5.2 on its internal Z.ai Code Bench, and it reports results on several public coding benchmarks. Z.ai also reports an 84.5 score on CyberGym and describes the model as state of the art for vulnerability discovery. The supplied research found no independent validation of those coding or cybersecurity claims. AWS says Z.ai did not report a direct performance comparison between GLM 5.3 and GLM 5.