GLM 5.1 by Z.ai

GLM 5.1 is a large language model built by Z.ai, a Chinese AI company. It processes text only, no images, and takes in about 200,000 tokens per request. A token is roughly half a word, so 200,000 tokens are several hundred pages of text. The weights, meaning the trained model file, have been public on Hugging Face under the MIT license since 3 April 2026. Anyone may download, modify, redistribute and commercially use them.

At lowcloud, GLM 5.1 is one of 14 models that run on servers in the EU. You pick it per message in chat, fix it for one agent, or set it for a single step of a workflow. The request does not go to Z.ai in China. It goes through Melious AI GmbH in Saarbrücken to data centres in the EU. One thing up front: GLM 5.1 is two generations old. For most jobs, GLM 5.3 or GLM 5.2 is the better choice today. There is exactly one reason to still pick GLM 5.1, and it is further down this page.

Vendor: Z.ai, China
Context window: 200k Token
Reads images: no
Licence: MIT
Processing: EU servers via Melious

Where GLM 5.1 comes from

The maker is Z.ai, an AI company in China. The model belongs to the GLM 5 series. The weights appeared on Hugging Face on 3 April 2026. Z.ai does not state an official announcement date in its documentation, so the Hugging Face date is the earliest one that can be verified. The license is MIT. That is one of the most permissive licenses there is: download, modification, redistribution and commercial use are allowed, with no meaningful conditions attached. So for GLM 5.1 the common label "open source" is accurate. For other models in the same series it is not: the successor GLM 5.3 is published under a Z.ai licence of its own, not under MIT. Technically, GLM 5.1 is a Mixture of Experts model, or MoE. That means the model is split into many sub-blocks and only a small share of them is used for each token. In total GLM 5.1 has about 754 billion parameters, the trained numerical values inside the model. Per token it runs 8 of 256 expert blocks plus one fixed block. That is why running it costs less than the total size suggests. These figures come from the configuration file of the weights on Hugging Face.

Two generations old: what that means

GLM 5.1 was followed by GLM 5.2 (weights published June 2026) and GLM 5.3 (weights published August 2026). Per the lowcloud catalogue, both hold 1 million tokens instead of 200,000. Usage of GLM 5.1 is around 2.6 percent of the usage of GLM 5.2. In practice the model has been superseded. If you are building a new agent today, take GLM 5.3 or GLM 5.2. They are newer, have a larger context window, and Z.ai keeps working on them. This page is not a recommendation for GLM 5.1. It describes what the model is and the one case in which it can still be the right pick.

What "on EU servers" actually means

The chain is: you, then lowcloud UG in Bielefeld, then Melious AI GmbH in Saarbrücken, then data centres in the EU. Melious runs no data centres of its own. It distributes requests across European operators. Your request does not go to Z.ai and not to a server in China. The model is Chinese, the operation is not. What we do not know, and therefore do not claim: which specific operator serves GLM 5.1 at a given moment. Melious does not publish which model runs in which data centre. If your client needs that detail, ask us before you commit to it.

What GLM 5.1 is good for

The one reason to use GLM 5.1 is fixed JSON schemas. A fixed JSON schema means you tell the model in advance exactly which fields the answer must contain, for example supplier, amount and date, and you reliably get the answer back in that shape. You need this wherever the answer feeds straight into another system: a CRM, accounting, a database, a spreadsheet. We tested this ourselves against the API instead of copying it from the maker's page: GLM 5.1 accepts fixed JSON schemas. Its successors GLM 5.2 and GLM 5.3 do not, in our tests, even though Z.ai's documentation lists "structured output" for both. That is our own observation. We know of no other source that publishes it. Within the GLM series, that makes GLM 5.1 the model for steps that need a fixed answer format. On top of that, 200,000 tokens are enough for long documents, contracts or whole email threads in one pass. For plain text work the model is usable: summarising, sorting, pulling out single fields, rewriting text.

What GLM 5.1 is not good for

GLM 5.1 cannot read images. Not poorly, but not at all: the model takes text only. Z.ai's documentation says so, and the configuration file of the weights confirms it, since it contains no image component. Scanned invoices, screenshots, photos and forms that sit as an image inside a PDF will not work. From the EU catalogue, these models read images: GLM 5.3 Flash, Kimi K3, Kimi K2.7 Code, Kimi K2.6, Qwen 3.8 27B and Mistral Small 4. If you need image input and fixed JSON schemas together, Kimi K2.6 is the only model in the EU catalogue that does both. For very long inputs GLM 5.1 is not enough: 200,000 tokens are far below GLM 5.2, GLM 5.3, Kimi K3 and DeepSeek V4 Pro at 1 million each. And for anything that just needs to be current and generally strong, GLM 5.1 is simply the older model. Take the successor instead.

What GLM 5.1 costs

We do not publish prices on this site. A margin is added to the purchase price, so any number printed here would be wrong. The relative placement: within the EU catalogue GLM 5.1 sits in the middle on price. It is neither the cheapest nor the most expensive of the 14 models. The actual amount in euros is shown on every single run in the app. That applies to all models equally, so you can run two models before deciding and compare the amounts. To choose a model deliberately you need an account. Without one you can have a workflow built at agent.lowcloud.de, and lowcloud picks the model for you. Model choice comes with the free account.

Try GLM 5.1

Picking a model needs a free account. You then choose GLM 5.1 in the chat, per agent, or per step.

Book a call

Common questions

Is GLM 5.1 open source?

Yes, in the narrow sense. The weights are on Hugging Face under the MIT license. Anyone may download, modify and commercially use them. Training data and training code are not included, which is the case for almost all so-called open models. At lowcloud you do not use GLM 5.1 as a download but as a service running on servers in the EU.

Does my data go to China?

No. The model comes from China, the processing does not. Your request runs from lowcloud in Bielefeld through Melious AI GmbH in Saarbrücken to data centres in the EU. Z.ai does not see the request and takes no part in the processing. What we cannot tell you: which European operator executes an individual run. Melious does not publish that assignment.

Can GLM 5.1 read images or scanned PDFs?

No, GLM 5.1 handles text only. For images, screenshots and scanned documents, take GLM 5.3 Flash, Kimi K3, Kimi K2.7 Code, Kimi K2.6, Qwen 3.8 27B or Mistral Small 4 from the EU catalogue.

GLM 5.1 or GLM 5.2 / GLM 5.3?

For new agents, GLM 5.3 or GLM 5.2. They are newer and hold 1 million tokens instead of 200,000. Pick GLM 5.1 only if your step needs a fixed JSON schema: in our own tests the two successors do not accept one, while GLM 5.1 does.

How large is the context window of GLM 5.1?

About 200,000 tokens for the input. The context window is the amount of text the model can see at once in a single request, meaning your question, the attachments and the conversation so far combined. 200,000 tokens are several hundred pages. The model file states the exact figure of 202,752 tokens.

What does one run with GLM 5.1 cost?

We do not publish prices, because a margin is added to the purchase price. Within the EU catalogue GLM 5.1 sits in the middle on price. The actual amount in euros is shown on every run in the app.

GLM 5.1 is a product name of Z.ai. There is no business relationship between lowcloud and Z.ai; we run the openly available model through a European provider. Last updated: 7 September 2026.