Amazon Bedrock now offers GLM 5.3, a model from Z.ai (Zhipu AI), AWS said on 5 October. AWS calls it a 753B-parameter mixture-of-experts model for coding and long agent tasks. That design splits the model into specialist parts, and only some work on each request. Customers use managed APIs, so they run no servers. It supports prompt caching (reusing repeated input to cut cost and delay) and Flex, Standard and Priority tiers.
Only eligible enterprise customers get access, and AWS did not say what qualifies or give prices. Customers pick the Region they send requests to, and Bedrock routes each request on for processing. The performance claims come from Z.ai: a 50% gain over GLM 5.2 on its own internal coding test, and 84.5 on the CyberGym security benchmark. AWS said Z.ai gave no direct comparison with GLM 5.
This turns an open-weight model into a managed service, so the buying questions become eligibility, tier prices and where routed requests are processed, which matters under data-residency rules. AWS says that in its setup for Strix, an open-source AI security-testing agent, inference runs under the customer's AWS account controls. The coding and security benchmark claims are Z.ai's own, so test them first.