Z.aiDedicated capacity
GLM-5.3 on capacity reserved for you
GLM-5.3 by Z.ai is an open-weight mixture-of-experts model with about 750B parameters and a context of 1M tokens. We run it for you on purpose-built AI chips in our data center in Munich, on capacity no one else uses. Lead time on request.
What GLM-5.3 is built for
Z.ai built GLM-5.3 on the base model of GLM-5.2 and improved it in post-training for complex coding and long tasks. It takes text input, and it is an open-weight model.
Agentic coding
Coding work over many steps: agents that read a codebase, change many files and run the tests. Z.ai reports its results in common coding-agent tools.
Long tasks
Tasks that run over many steps and tool calls, where the model has to keep the goal and the current state in view.
1M tokens of context
A whole codebase, a large set of contracts or a long agent history fits in one request, without splitting it into pieces first.
Tools and structured output
Function calling and structured output, so an agent can call your tools and hand back data your systems can read.
Open weights
Z.ai publishes the weights under its own GLM-5.3 licence. It is permissive; only Model-as-a-Service companies with more than USD 10B revenue need a security review by Z.ai first.
Why GLM-5.3 runs on dedicated capacity
Runs in our data center in Munich on dedicated capacity. GLM-5.3 is not on the Inference API: a model of this size takes a large part of a rack's memory, so we reserve the capacity for one customer and size it to their workload.
Guaranteed throughput
No other tenants and no shared queue. Your requests get the capacity you reserved.
Run by us
We deploy, operate and monitor the model for you, with ISO 27001 certified operations.
Predictable monthly cost
A flat monthly cost for the capacity reserved for you, not a bill per token.
Questions about GLM-5.3
Can I call GLM-5.3 on the Inference API?
No. GLM-5.3 runs only on dedicated capacity. On the Inference API, our EU sovereign models are gpt-oss-120b and Gemma 4.
How large is GLM-5.3, and how much context does it take?
About 750B parameters in a mixture of experts, with about 40B of them active per token, and a context window of 1M tokens. It takes text input.
How fast is GLM-5.3 on Infercom?
We publish a speed only after we have measured it ourselves, and we have not published one for GLM-5.3 yet. For a sizing, we measure against your own prompt lengths and output ratio.
Which licence applies to GLM-5.3?
Z.ai's own GLM-5.3 licence. It allows commercial use. Only Model-as-a-Service companies with more than USD 10B revenue in twelve months need a security review by Z.ai before they use it commercially.
How long until GLM-5.3 runs for us?
Lead time on request. It depends on the capacity your workload needs - tell us about it and we'll give you a date.
Does my data stay in the EU?
On dedicated capacity, GLM-5.3 runs on our own hardware in our data center in Munich, and prompts and outputs are not persisted. SambaNova, the US company that makes our purpose-built AI chips, operates the software stack with remote access; our data processing agreement names it.
Can we test GLM-5.3 before we commit?
Yes. We can arrange a time-limited test on dedicated capacity before you sign, so you see how GLM-5.3 handles your own workload.