GLM 5.3 Flash

GLM 5.3 Flash is an AI language model from the Chinese company Z.ai, released on 26 August 2026. It is published under the MIT licence, the weights are public on Hugging Face, and it reads images as well as text. It has 320 billion parameters, of which roughly 18 billion are used per request. It is the smaller, faster variant alongside the larger GLM 5.3 of the same generation.

At lowcloud, GLM 5.3 Flash is one of 14 models that run through Melious AI GmbH on servers inside 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; it goes to a copy of the model running on European hardware.

Vendor: Z.ai, China
Context window: 1 Mio. Token
Reads images: yes
Licence: MIT
Processing: EU servers via Melious

Where the model comes from

The maker is Z.ai, based in China. The model was released on 26 August 2026. The licence is MIT. That is one of the most permissive licences there is: commercial use allowed, no extra conditions, no cap on user numbers. For Chinese models this is not the norm; many ship under a custom licence with conditions attached. Here the term "open source" actually holds: the weights, meaning the model's trained numerical values, are public on Hugging Face under zai-org/GLM-5.3-Flash, along with a variant at a different numerical precision (BF16). Technically it is an MoE model. MoE stands for Mixture of Experts: the model is split into many sub-networks, and only a small share of them runs per request. 320 billion parameters in total, around 18 billion active per request. Hence the Flash label. Z.ai describes the architecture as a mix of sparse and linear attention; that is the maker's own statement and we have not verified it. The context window figures do not agree. The context window is how much text the model can read at once in a single request, measured in tokens; a token is roughly half a word to a whole word. The lowcloud catalogue and Z.ai's documentation both state 1 million tokens. The model card on Hugging Face names 300,000 tokens as the maximum used in its evaluation. At what length answer quality drops off in practice, we have not measured. There is no successor as of 7 September 2026. GLM 5.3 Flash is the current state of the GLM line, alongside the larger GLM 5.3.

What "on EU servers" actually means here

The request passes through three stations: from you to lowcloud UG in Bielefeld, from there to Melious AI GmbH in Saarbrücken, from there into a data centre in the EU. Melious runs no data centres of its own. It distributes requests across European operators. Which operator serves a given model is something Melious does not publish. So we do not know it either and cannot tell you. If your client asks for that detail, this is the point where the chain ends for us. What is settled: the request does not go to Z.ai. Because the weights are freely available, the model runs as a copy on European hardware. The maker does not see your inputs.

How it differs from GLM 5.3

GLM 5.3 is the larger model of the same generation, 753 billion parameters. It reads text only. GLM 5.3 Flash is smaller, faster, and the only model in the GLM line in the lowcloud catalogue that reads images. If you want to stay inside the GLM line and need to process images, it is the only option. Z.ai calls it the first natively multimodal model of the GLM 5 line, meaning image handling was part of the design rather than bolted on later. Neither of the two accepts fixed JSON schemas. More on that below.

What the model is good for in a workflow

Reading images. Invoices, delivery notes, screenshots, scans, photos of filled-in forms. The model states what is on them, and the next step of the workflow carries on with that text. Long inputs. A contract, a set of minutes, a full email thread all fit into one request without you having to split the text first. Steps that run often. Because only a small share of the model runs per request, it suits workflows that execute many times a day: sorting incoming messages, checking subject lines, pulling fields out of a text, drafting replies. Tool calls. The model can trigger functions, so it can set off a step in a workflow instead of only returning text. Everyday text work. Summarising, classifying, rewriting, translating.

What it is not good for

Fixed JSON schemas. A fixed JSON schema is a template that pins down field names and data types so the next step can rely on the structure. GLM 5.3 Flash does not accept one: the provider rejects the request with an error. That is our own observation from probing the API, not a statement by the maker. Z.ai's documentation lists "structured output (JSON)" as a capability, but that refers to the loose form, where the model is asked for JSON without a schema being enforced. If a step strictly needs a fixed field structure, pick another model from the EU catalogue: Kimi K2.6 (256k context, reads images), GLM 5.1 or GLM 5 (200k context, no images), Qwen 3.5 397B (262k context, no images), or DeepSeek V4 Flash. You can mix models step by step inside one workflow: read the images with GLM 5.3 Flash, then hand the result to Kimi K2.6 to force it into fixed fields. Clients who will not accept a model developed in China. This comes up in tenders and internal policies. In the lowcloud EU catalogue, Mistral Small 4 (Mistral AI, France) is the only non-Chinese model; it reads images as well. Work where you want the strongest model available. Z.ai positions Flash explicitly as the cheaper and faster variant next to GLM 5.3, not as the more capable one.

What it costs

We publish no figures here. A markup is applied to the purchase price, so any number printed on this page would be wrong. The placement: GLM 5.3 Flash sits in the second-cheapest price tier of the lowcloud EU catalogue. What a run actually cost is shown in euros on every single run in the app. So you see per workflow what it cost you and can pass that on to your client. Without an account you can have a workflow built for you at agent.lowcloud.de; in that case lowcloud picks the model. Free model choice comes with the free account.

Try GLM 5.3 Flash

Picking a model needs a free account. You then choose GLM 5.3 Flash in the chat, per agent, or per step.

Book a call

Common questions

What is GLM 5.3 Flash?

An AI language model from the Chinese company Z.ai, released on 26 August 2026 under the MIT licence. 320 billion parameters in total, around 18 billion active per request. It handles text and images and is the smaller, faster variant next to GLM 5.3.

Does my data go to China if I use GLM 5.3 Flash?

No. The model weights are freely available, so lowcloud runs a copy on servers inside the EU. The path is: you to lowcloud UG in Bielefeld, from there to Melious AI GmbH in Saarbrücken, from there into a data centre in the EU. Z.ai is not part of that path and does not see your inputs. What we do not know: which European operator serves the model specifically. Melious does not publish that.

Is GLM 5.3 Flash open source?

The licence is MIT, a genuinely free licence with no extra conditions, and the weights are public on Hugging Face. For Chinese models that is the exception; many use a custom licence with conditions. The training data is not disclosed. What is open are the finished weights.

Can GLM 5.3 Flash read images?

Yes. It is the only model in the GLM line in the lowcloud EU catalogue that processes images. The larger GLM 5.3, as well as GLM 5.2, GLM 5.1 and GLM 5, take text only. If a workflow needs image handling and should stay inside the GLM line, GLM 5.3 Flash is the answer.

Why does GLM 5.3 Flash reject my JSON schema?

Because the provider does not accept fixed JSON schemas and returns an error for the request. We probed this against the API ourselves. A loose "answer in JSON" works; an enforced field structure does not. For steps that need a fixed schema, use Kimi K2.6, GLM 5.1, GLM 5, Qwen 3.5 397B or DeepSeek V4 Flash.

What does GLM 5.3 Flash cost at lowcloud?

We do not publish prices, because a markup is applied to the purchase price. Within the EU catalogue, GLM 5.3 Flash sits in the second-cheapest price tier. The actual cost in euros is shown on every single run in the app.

GLM 5.3 Flash 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.