GLM 5.3 from Z.ai arrives on Amazon Bedrock as a managed API

A 753B-parameter model for coding and agents

GLM 5.3, as published on Hugging Face Hub, is a 753B-parameter mixture-of-experts model optimized for coding and long-horizon agentic tasks. AWS frames the release around the growing demands of such workloads: refactoring a repository spanning hundreds of files, sustaining a multi-hour agentic workflow without losing context, and reasoning through complex systems problems with tool use at every step.

Running open-weight models for these jobs has historically meant provisioning and operating your own inference infrastructure, AWS says. On Bedrock, the model is offered through managed APIs instead.

Z.ai has also reported that the model shows notable cyber security capabilities. According to the company, it measured a leading score of 84.5 on the CyberGym benchmark at release.

What changes compared to GLM 5

GLM 5 arrived on Amazon Bedrock earlier this year, and AWS describes GLM 5.3 as building on the same lineage. The post lists three main gains, all based on Z.ai's own reporting:

• Stronger coding: Z.ai claims competitive performance on coding benchmarks including DeepSWE, Terminal Bench 3.0 and FrontierSWE. It also reports a 50% improvement over GLM 5.2 on its own internal coding benchmark. Direct comparisons to GLM 5 were not reported, because the size of the improvements led to updating the benchmark tests themselves since the GLM 5.1 announcement.
• Emergent cyber security capabilities: The reported security results make the model a natural fit for defensive security workflows, according to the post.
• Broader Bedrock integration: Cross-Region inference profiles, implicit and explicit prompt caching, and improved feature parity of the OpenAI-compatible Responses and Chat Completions APIs alongside Invoke and Converse.

API access, caching and service tiers

Developers can call GLM 5.3 through the OpenAI-compatible Responses and Chat Completions APIs, or through the Amazon Bedrock Invoke and Converse APIs. For new applications, AWS recommends the OpenAI-compatible APIs because they support a more complete set of features. The model can also be tried without code in the Bedrock console, under Test > Playground.

Prompt caching is supported in two modes. Implicit caching is on by default. Explicit caching, which AWS recommends on the Responses and Chat Completions APIs, lets developers mark reusable prompt prefixes with prompt_cache_breakpoint markers, enabled via prompt_cache_options. Each breakpoint must contain at least 1,024 tokens to be eligible. AWS says this can further improve cache hit rate, and therefore latency and cost savings, for agentic workloads that resend large system prompts or repository context every turn.

Cross-Region inference is available through a US profile (us.zai.glm-5.3) and a Global profile (global.zai.glm-5.3). Requests go to a chosen source AWS Region, and Bedrock routes each one for processing. Three service tiers are offered: Flex for less time-sensitive workloads, Priority for latency-critical requests at a higher price, and Standard as the default balance between price and speed.

On credentials, AWS says it strongly recommends short-lived credentials over long-lived API keys where possible. Its Python example uses the OpenAI SDK with the aws-bedrock-token-generator library to create short-term tokens from standard AWS CLI credentials.

Security testing with Strix

The post's showcase workload is authorized security testing with Strix, an open-source AI penetration testing agent that runs code dynamically, finds vulnerabilities and validates them with proof-of-concept tests. According to AWS, the Strix documentation uses GLM 5.3 as its default model at the time of writing. Strix can be configured to use GLM 5.3 on Bedrock instead of a third-party inference provider, so inference runs under the AWS account's controls.

AWS stresses that users should only test applications they own or have explicit written permission to test, and notes that unauthorized testing is illegal in most jurisdictions and violates the AWS Acceptable Use Policy. The walkthrough targets OWASP Juice Shop, a deliberately vulnerable sample application running locally. For managed, continuous testing, AWS points to AWS Continuum, which it describes as complementary to open-source agents.

What it means for users

For enterprise teams that already work on AWS, the release offers a large open-weight coding and agent model behind managed APIs, with caching, cross-Region routing and pricing tiers built in. Access is limited to eligible enterprise customers, and the performance claims come from Z.ai rather than independent testing.