GLM-5.3 is now live on Amazon Bedrock. On October 6, 2026, Zhipu's Z.ai announced that its flagship model had joined AWS's managed model platform; AWS's official documentation lists the model's launch date as October 5 and notes that access is limited to eligible customers. According to a report the same day by the STAR Market Daily, AWS will share revenue with Zhipu based on model usage — meaning Zhipu's overseas income will increasingly follow each API call rather than one-off licensing deals.
The model itself: 744B parameters aimed at coding agents
AWS's official model page describes GLM-5.3 as Zhipu's flagship open-weight text model. It uses a mixture-of-experts architecture with 744B total parameters and roughly 40B active per token, built on the same base as GLM-5.2 with improvements from scaled post-training. It offers a 1M-token context window, up to 128K output tokens, and selectable reasoning-effort levels so callers can trade latency and token consumption against task quality.
The listed use cases make its target unmistakable: agentic coding and autonomous software engineering, terminal and CLI tasks, repository-scale code generation and refactoring, and long-horizon engineering work spanning many tool calls. Function calling, structured JSON output, and streaming — including streamed reasoning — are all supported. This model is not competing on small talk; it is competing for a place inside enterprise development pipelines.
The revenue share is the real story
Listing a model on a major cloud is not new. How the money splits is. A usage-based revenue share turns the model maker into a supplier on the cloud's shelf: the cloud provides global regions, compliance cover, enterprise relationships, and billing, while the model maker provides capability and takes a cut of usage. The report says Zhipu has recently signed similar arrangements with several overseas cloud providers; at home, Alibaba Cloud's Bailian platform has a comparable deal, and Huawei Cloud already lists GLM-5.3 with a partnership intended. A revenue-share network spanning domestic and overseas clouds is replacing the old one-by-one licensing grind.
For Zhipu, the upside is a light footprint: no sales team in every country, and revenue that scales with real usage. The cost is that pricing power and the customer relationship sit with the cloud provider, and the split itself is undisclosed. For AWS, adding a competitive open-weight flagship keeps enterprise customers from drifting to rival clouds.
Two gates before you can use it
First, eligibility. AWS states plainly that GLM-5.3 is available only to eligible customers, who should contact their AWS account team — this is not a sign-up-and-call model. Second, deployment shape: the model is offered through cross-Region inference only, via the US geographic inference profile (us.zai.glm-5.3) or the global profile (global.zai.glm-5.3), with no in-Region on-demand option. For enterprises with strict data-residency rules, the possibility of cross-border requests needs assessing up front; that is a separate question from how strong the model is.
In the bigger picture, Chinese open-weight models are shifting from winning leaderboards to earning money on global cloud shelves. Benchmarks decide who gets looked at; revenue-share channels decide whether R&D pays for itself. Whether this move works will not be settled by launch-day noise, but by real Bedrock usage over the coming quarters.
