Z.ai's GLM 5.3, a 753B open-weight MoE, goes GA on Amazon Bedrock

Amazon Bedrock has made GLM 5.3 from Z.ai generally available. The model is a mixture-of-experts with 753 billion total parameters and roughly 40 billion active per token, designed for agentic coding and long-horizon software engineering.

It can be called via OpenAI-compatible APIs, which lowers switching cost for teams already on OpenAI SDKs, and supports prompt caching to cut cost and latency. AWS demonstrates it running an authorized security test using the open-source Strix agent. Mistral also moved GLM 5.3 to GA on its platform.

The launch adds another strong Chinese open-weight model to a hyperscaler catalog, alongside the DeepSeek LiveBench news this week. For enterprises chasing cheaper coding agents after high frontier bills, a large open-weight option inside AWS governance is attractive.

Watch pricing versus Claude and GPT on Bedrock, and whether regulated customers accept a Chinese-origin model for production work.