Skip to content
NewsProduct launch

Amazon Bedrock adds Z.ai's GLM 5.3 as a managed coding model option

· by Pondero Newsdesk

The short version

AWS made Z.ai's 753-billion-parameter GLM 5.3 generally available on Amazon Bedrock on October 5, giving regulated buyers a procurement path to a Chinese open-weight coding model without a direct vendor contract.

Amazon Bedrock adds Z.ai's GLM 5.3 as a managed coding model option

Enterprises that cannot easily sign a direct contract with a Chinese AI lab now have a procurement shortcut to one of its frontier coding models. Amazon Bedrock listed general availability for Z.ai's GLM 5.3 on October 5, 2026, per AWS's announcement, making it the fourth Z.ai model Bedrock now carries in its catalog alongside GLM 4.7, GLM 4.7 Flash, and GLM 5, per AWS's Bedrock documentation.

What

GLM 5.3 is a mixture-of-experts model with roughly 753 billion total parameters and about 40 billion active per token, per AWS's announcement. On Bedrock it ships with a 1-million-token context window, up to 128K output tokens, always-on reasoning with selectable effort levels, and explicit prompt caching that lets customers reuse cached system prompts and messages to cut latency and input cost. AWS is pitching the listing at agentic coding and long-horizon software-engineering work, available to eligible enterprise customers through US and Global cross-Region inference profiles.

The model is not new. Z.ai released GLM 5.3 on August 14, 2026, built on the same base architecture as GLM 5.2, with every performance gain coming from additional post-training rather than a new model, per Z.ai's own blog post. Z.ai said the result improved 50% over GLM 5.2 on its in-house Code Bench and reached open-weight state of the art on CyberGym, a benchmark for finding software vulnerabilities, while also posting large jumps on exploitation benchmarks that the company called an "emergent cyber capability." AWS's own post does not repeat any of those performance numbers; it describes only the architecture, context window, and Bedrock plumbing.

A compliance shortcut, not a benchmark verdict

For a team bound by procurement rules, data-residency policy, or a vendor-risk review that makes contracting directly with a Chinese lab impractical, Bedrock's listing turns GLM 5.3 into a line item inside existing AWS billing, IAM, and region controls rather than a new vendor relationship. That is the actual shift this GA unlocks: access, not evaluation. Buyers who want to confirm Z.ai's coding or vulnerability-detection claims still have to go to Z.ai's own benchmark post or wait for independent replication, since AWS's announcement does not vouch for those specific numbers.

What to watch next

Whether Z.ai's self-reported CyberGym and Code Bench results get independent testing now that the model sits inside a mainstream US cloud's catalog, and whether Microsoft Azure or Google Cloud add their own GLM 5.3 listings in response.

Sources