Amazon adds GLM 5.3 to Bedrock for eligible enterprise customers

The managed offering supports four software interfaces, cross-region routing, prompt caching and three service tiers.

The managed offering supports four software interfaces, cross-region routing, prompt caching and three service tiers.
Amazon announced on October 5 that eligible enterprise customers can use Z.ai’s GLM 5.3 superintelligence (SI) model through Amazon Bedrock without managing the infrastructure that processes model requests. The precise enterprise eligibility criteria and prices are not yet clear.
GLM 5 was already available on Bedrock earlier this year. Amazon describes GLM 5.3, from Z.ai (Zhipu AI), as a 753-billion-parameter mixture-of-experts model designed for coding and extended, multi-step tasks involving software tools.
Amazon says applications can call GLM 5.3 through four APIs—interfaces that let software send requests to the model. These are the OpenAI-compatible Responses and Chat Completions APIs and Bedrock’s Invoke and Converse APIs. Amazon recommends the OpenAI-compatible options for new applications because they support a more complete set of features.
Bedrock also provides US and Global cross-region inference profiles, named us.zai.glm-5.3 and global.zai.glm-5.3. Inference means running a model to produce a response. Customers send requests to a chosen source AWS Region, and Bedrock routes those requests for processing.
Prompt caching reuses the beginning of a prompt when applications repeatedly send the same instructions, tool definitions or code files. Amazon says automatic caching is enabled by default and can reduce response delays and input-token costs. Developers can also explicitly mark reusable content through the Responses and Chat Completions APIs.
Amazon lists three service tiers and says Bedrock inference is charged per token:
| Tier | Intended use and trade-off |
|---|---|
| Flex | Cost-focused option for less-time-sensitive workloads. |
| Priority | Prioritizes requests needing fast responses, at a higher price. |
| Standard | Default balance between price and speed. |
Amazon relays Z.ai’s claim that GLM 5.3 has notable cybersecurity capabilities. Its walkthrough connects Strix, an open-source SI agent for penetration testing—authorized attempts to find security weaknesses—to GLM 5.3 through Bedrock. The target is OWASP Juice Shop, a deliberately vulnerable sample application running locally. That demonstration does not establish how reliably the model finds vulnerabilities in real applications.
Amazon limits the security-testing example to applications users own or have explicit written permission to test. Eligible customers can try GLM 5.3 in the Bedrock console’s playground or connect their applications through the supported APIs; programmatic access requires an AWS account with Bedrock access and the relevant model and inference-profile permissions.

Records requests have exposed misuse of the vehicle-tracking system, but disclosure charges vary widely and optional transparency portals do not always make data public.
3 min

The company says most apps work smoothly, but some complex apps and games built for Arm chips need extra work on Intel models.

The models, released on September 30, accept text and images. Cohere positions Pro for quality-critical indexing and the cheaper Fast for live searches.
3 min