Amazon adds Z.ai's 753B GLM 5.3 coding model to Bedrock
Amazon says Z.ai's 753-billion-parameter coding model is open to eligible enterprise customers on Bedrock, and Z.ai claims a 50 percent gain over GLM 5.2 on its own test.
Amazon Web Services has added GLM 5.3, a model from Z.ai, to Amazon Bedrock, according to an AWS Machine Learning blog post published on 5 October. AWS describes it as a 753-billion-parameter mixture-of-experts model tuned for coding and long-horizon agentic tasks. The post says the weights are also on the Hugging Face Hub. No independent testing of the Bedrock version has been published yet.
Z.ai reports competitive results on coding benchmarks including DeepSWE, Terminal Bench 3.0 and FrontierSWE, the post says. The company also claims a 50 percent improvement over GLM 5.2 on its own internal coding benchmark, and a score of 84.5 on the CyberGym benchmark at release, which it calls leading. All of those are figures from Z.ai, and the post does not publish the underlying scores.
On the platform side, Bedrock offers prompt caching in implicit and explicit modes, and cross-Region inference through US and Global profiles. Customers can pick a Flex, Priority or Standard service tier. The model can be called through either an OpenAI-compatible API or the native Bedrock APIs, which could make it easier to swap into tools that already speak the OpenAI format.
Access is currently limited to eligible enterprise customers, and the post says billing is pay-per-token. It does not give per-token prices, the context window or the licence terms, so buyers would need to check those separately. AWS did not say when the model would reach other customers, and the post does not name the Regions where the profiles route requests.