GLM 5 by Z.ai
GLM 5 is a language model from the Chinese company Z.ai, based in Beijing, released in February 2026. The technical report was published on 17 February 2026. The model has 744 billion parameters, of which around 40 billion are active per request. It reads a context window of 200,000 tokens, it handles text only and cannot read images, and its weights are publicly downloadable under the MIT license. Parameters are the trained numbers inside the model. Tokens are the units text is split into; one token is roughly half a word to a whole word. The context window is how much text the model can read at once in a single request.
On lowcloud, GLM 5 is one of 14 models that run through Melious AI GmbH in Saarbruecken on servers inside the EU. The request never reaches Z.ai. You choose which model works: per message in chat, fixed for one agent, or for a single step of a workflow. GLM 5 is the oldest GLM version in the lowcloud catalogue. Z.ai has since released GLM 5.1, GLM 5.2, GLM 5.3 and GLM 5.3 Flash. There are only a few reasons left to pick GLM 5 for a new agent. Those reasons are listed below.
- Vendor
- Z.ai, China
- Context window
- 200k Token
- Reads images
- no
- Licence
- MIT
- Processing
- EU servers via Melious
Where GLM 5 comes from
The maker is Z.ai, a company based in Beijing, China. The model was released in February 2026. The architecture is a Mixture of Experts, or MoE. The model is split into many sub-sections and switches on only a small share of them per request: 40 of 744 billion parameters. That lowers the cost of running it. It changes nothing about how you use it in the builder. The license is MIT. That is a genuine open source license with no conditions on how the model may be used and no carve-outs for particular customer types or revenue thresholds. The weights are public on Hugging Face at zai-org/GLM-5, so anyone can download them and run the model themselves. This is worth stating because it no longer holds for the newer GLM versions. On Hugging Face, GLM 5.3 is published under a custom Z.ai license named glm-5.3, not under MIT. If you need to prove to a client that a model is genuinely freely licensed, you can make that case for GLM 5 and not for GLM 5.3.
GLM 5 has been superseded
Z.ai has released four further versions since GLM 5: GLM 5.1, GLM 5.2, GLM 5.3 and GLM 5.3 Flash. All four are available on lowcloud as well. The current version is GLM 5.3. The most visible difference is the context window. GLM 5 holds 200,000 tokens, GLM 5.2 and GLM 5.3 hold one million. That is the difference between a long contract and a full folder of files in one request. Two cases still favour GLM 5. First, an existing agent already runs on it and delivers what it should. There is no reason to change a workflow that works. Second, you need fixed JSON schemas. A JSON schema is a template that states exactly which fields the answer must contain and what type each field has. GLM 5 accepts such templates; according to the lowcloud catalogue, GLM 5.2 and GLM 5.3 do not. Within the GLM line that leaves GLM 5 and GLM 5.1.
What "runs on EU servers" actually means here
The chain is short and has three stations. Your request goes to lowcloud UG in Bielefeld. From there it goes to Melious AI GmbH in Saarbruecken. Melious passes it to a data centre inside the EU where GLM 5 runs. Z.ai is not part of this chain. The model was developed and trained in China. The copy that handles your request runs in Europe. The maker never sees the request. Melious does not operate its own data centres; it distributes load across European operators. Melious does not publish which operator serves a given model. lowcloud therefore does not know either and cannot tell you. If a client needs the specific location written into a contract, that is the limit of what you can state.
What GLM 5 is good for in the builder
Text work with a fixed output format. GLM 5 accepts fixed JSON schemas, so the answer comes back in the structure you defined. That matters whenever a later step in the workflow has to work with the data: pulling fields out of an email and writing them into a CRM, moving line items from invoice text into a spreadsheet, sorting incoming requests into fixed categories. Without a schema the next step has to guess. With one it does not. lowcloud tested this against the API rather than copying it from the vendor page. Here the result matches what Z.ai states: fixed schemas are accepted. Documents up to roughly 200,000 tokens. That covers long contracts, manuals, or several files bundled into one request. Steps that should keep running unchanged and need no new abilities. An agent that has worked on GLM 5 for months does not automatically get better from a model swap, and it will need retesting afterwards.
What GLM 5 is not good for
Images. GLM 5 cannot read them. Screenshots, photographed delivery notes, scanned forms, product photos: all of these need a different model. lowcloud tested this against the API too, and the result matches what Z.ai states. If you need image reading and fixed JSON schemas at the same time, Kimi K2.6 is the only model in the EU catalogue that does both. If only the image matters and the output format can stay free, GLM 5.3 Flash, Mistral Small 4, Qwen 3.8 27B, Kimi K3 and Kimi K2.7 Code are the options. Very long inputs. GLM 5 stops at about 200,000 tokens. For one million tokens there are GLM 5.2, GLM 5.3, Kimi K3 and DeepSeek V4 Pro. Tasks where you want the current state of the art. For coding work and long multi-step workflows, GLM 5.3 is the newer work from the same maker. lowcloud publishes no comparison figures for this, and a vendor's own benchmarks are not an independent measurement.
What GLM 5 costs
lowcloud does not publish token prices. The reason is straightforward: a margin is added on top of the purchase price, so any figure named here would be wrong. Within the EU catalogue, GLM 5 sits in the lower-middle price range. It is neither the cheapest of the 14 models nor the most expensive. What a run actually cost is shown in euros on every single run in the app. That is the number you bill your client from, not a list price on a website. To try it: on agent.lowcloud.de you can have a workflow built without an account, and lowcloud picks the model for you. Free model choice, GLM 5 included, comes with the free account.
Try GLM 5
Picking a model needs a free account. You then choose GLM 5 in the chat, per agent, or per step.
Common questions
Does my data go to China if I use GLM 5?
No. Your request goes to lowcloud UG in Bielefeld, from there to Melious AI GmbH in Saarbruecken, and from there to a data centre inside the EU. Z.ai never sees the request. The model comes from China; the copy serving you runs in Europe. What lowcloud cannot tell you is which European operator runs it. Melious does not publish that.
Is GLM 5 open source?
Yes. The weights are published under the MIT license on Hugging Face at zai-org/GLM-5. MIT puts no conditions on how the model is used. Be careful with overview sites that claim this for the whole GLM line: GLM 5.3 is published under a custom Z.ai license, not under MIT.
Can GLM 5 read images?
No. GLM 5 handles text only. lowcloud tested this against the API rather than taking the vendor's word for it. For images plus a fixed JSON schema, Kimi K2.6 is the only option in the EU catalogue. If only the image matters, GLM 5.3 Flash, Mistral Small 4, Qwen 3.8 27B, Kimi K3 and Kimi K2.7 Code also work.
How large is the context window of GLM 5?
200,000 tokens according to the lowcloud catalogue, which matches Z.ai's own documentation. That is roughly 150,000 words in one request. If that is not enough, GLM 5.2, GLM 5.3, Kimi K3 and DeepSeek V4 Pro offer one million tokens.
GLM 5 or GLM 5.3 - which should I use?
For new agents, GLM 5.3 in most cases: newer, and a larger context window. Pick GLM 5 only if you need fixed JSON schemas, which GLM 5.3 does not accept according to the catalogue, or if an existing agent already runs on GLM 5 without problems. Neither of them reads images.
What does a run with GLM 5 cost?
lowcloud names no token prices, because a margin is added on top of the purchase price and a published figure would therefore be wrong. Within the EU catalogue, GLM 5 sits in the lower-middle price range. The actual cost in euros is shown on every run in the app.
GLM 5 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.