GLM (Z.ai): the open-weight family Revolut chose for AI coding

Contents

1. What it does for a company
2. Published evidence
3. Strengths and limits
4. Pricing
5. Questions to ask before choosing GLM
6. Sources

Type	Open-weight model family
Headquarters	Beijing, China
Pricing	Published
Website	huggingface.co

GLM is the open-weight model family of Z.ai, a developer based in Beijing⁠1Z.ai: company siteZ.AI Co., Ltd.. It is the family Revolut moved its default AI coding model to in 2026, first GLM-5.1 and then GLM-5.3⁠2British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hostedThe Register · 3 Nov 2026.

Best for: AI coding at scale: the family Revolut moved about 2,000 engineers onto.

Not the right fit for: Companies whose policy rules out models from Chinese developers, even when self-hosted.

What it does for a company

The current flagship, GLM-5.3, is a 753-billion-parameter text model built on the same base as GLM-5.2⁠3zai-org/GLM-5.3Z.ai (Hugging Face). GLM-5.3-Flash is the smaller, multimodal option: 320 billion parameters in total, 18 billion active per token, with image and text input⁠4zai-org/GLM-5.3-FlashZ.ai (Hugging Face). GLM-5.2’s model card describes a one-million-token context⁠5zai-org/GLM-5.2Z.ai (Hugging Face). Z.ai documents serving with vLLM and SGLang, both of which expose an OpenAI-compatible interface⁠36zai-org/GLM-5.3Z.ai (Hugging Face)Online serving (OpenAI-compatible server)vLLM.

The licence depends on the version. GLM-5.3-Flash and GLM-5.2 are released under the MIT licence⁠45zai-org/GLM-5.3-FlashZ.ai (Hugging Face)zai-org/GLM-5.2Z.ai (Hugging Face). GLM-5.3 uses its own licence: it reads like MIT, but an operator offering the model as a service with more than $10 billion in group revenue in any 12 months must first pass a security review by Z.ai. There is no user cap and no naming rule⁠7GLM-5.3 LicenseZ.ai (Hugging Face).

The scale of the flagship sets the hardware. Revolut serves about 2,000 engineers from 56 Nvidia B200 GPUs, and GLM-5.1’s weights alone need roughly 380 GB of GPU memory⁠2British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hostedThe Register · 3 Nov 2026. A company that does not want to run that hardware can use per-token endpoints: Together AI lists GLM-5.3 at $1.40 per million input tokens and $4.40 per million output tokens⁠8Pricing: serverless inference, dedicated inference and GPU clustersTogether AI. Vendor-reported results for GLM-5.3 include 88.2 on Terminal Bench 2.1⁠3zai-org/GLM-5.3Z.ai (Hugging Face).

Published evidence

• Revolut. Revolut moved its default coding model to GLM-5.1 and then GLM-5.3 on 56 dedicated B200 GPUs and reports about 90% lower inference costs.⁠2British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hostedThe Register · 3 Nov 2026

Strengths and limits

Strengths

• The strongest public evidence of enterprise use: Revolut’s self-hosted deployment for about 2,000 engineers.
• GLM-5.3-Flash and GLM-5.2 are released under the MIT licence.
• Supported by vLLM and SGLang for OpenAI-compatible serving.

Limits

• GLM-5.3 itself uses a custom licence: a model-as-a-service operator with more than $10 billion in group revenue must pass a security review by Z.ai.
• The flagship needs a multi-GPU cluster; Revolut uses 56 B200s.
• We found no official fine-tuning guide on the model card.

Choose another option if: You need a model from a European or US developer, or one that runs on a single GPU.

Pricing

Weights are free. Together AI lists GLM-5.3 at $1.40 per million input tokens and $4.40 per million output tokens.

Questions to ask before choosing GLM

• Does our security policy permit open weights from a Chinese developer when they run on our own or a European provider’s hardware?
• Do we need GLM-5.3, or does the MIT-licensed GLM-5.3-Flash meet our quality bar at a fraction of the hardware?
• What volume of tokens per month makes dedicated GPUs cheaper than per-token endpoints for us?
• How will we fine-tune it on our own data, given that we found no official fine-tuning guide?

See how GLM compares with the other options in the main guide.

Sources

1. Z.ai: company site. Z.AI Co., Ltd.. Accessed 9 Nov 2026.
2. British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted. The Register, 3 Nov 2026. Accessed 9 Nov 2026.
3. zai-org/GLM-5.3. Z.ai (Hugging Face). Accessed 9 Nov 2026.
4. zai-org/GLM-5.3-Flash. Z.ai (Hugging Face). Accessed 9 Nov 2026.
5. zai-org/GLM-5.2. Z.ai (Hugging Face). Accessed 9 Nov 2026.
6. Online serving (OpenAI-compatible server). vLLM. Accessed 9 Nov 2026.
7. GLM-5.3 License. Z.ai (Hugging Face). Accessed 9 Nov 2026.
8. Pricing: serverless inference, dedicated inference and GPU clusters. Together AI. Accessed 9 Nov 2026.

Further reading

• Buyer's guide · Updated quarterlyThe best open-source AI in 2027, for companies that want to own itAn analysis of the open-weight model families a company can run today and the layer that turns one into its own, compared on seven published criteria, with sourced licences, costs and the established path from rented AI to owned AI.
• Case analysisGoing headless like Revolut: what replacing Claude Code with an open model tookRevolut now runs most of its AI coding on an open-weight model it serves itself, for about a tenth of its former bill. A reconstruction of what it built, what the numbers mean for other companies, and the step that comes after.
• Licence guideOpen source vs open weights: what each licence lets a company doAlmost every model sold as open-source AI publishes its weights but not its training data. The difference rarely matters for internal use. It matters a great deal for companies that resell a model or build a product on it.

Also in: Deutsch · Français · Español · Italiano · Nederlands · Polski