GLM (Z.ai)

What GLM (Z.ai) does

Z.ai, the Chinese lab formerly called Zhipu AI, builds the GLM family of language models. Its current flagship, GLM-5.3, is a 753B-parameter mixture-of-experts model with a 1M-token context window and up to 128K tokens of output. It is text-only, with reasoning always on. The weights have been on Hugging Face since late August 2026 under the "GLM-5.3 License". That licence is MIT-like for most uses, but Model-as-a-Service providers with more than US$10B in revenue need a Z.ai security review first. The smaller GLM-5.3-Flash is MIT-licensed. You can use the models through Z.ai's API, through its GLM Coding Plan inside tools like Claude Code, OpenCode, Kilo Code and Cline, or by hosting the weights yourself.

Good for

• Self-hosting a strong open-weight coding model on your own GPUs
• Cheaper coding-agent tokens through the GLM Coding Plan in Claude Code, OpenCode or Cline
• Teams that want a flat, predictable AI bill instead of per-seat frontier spend

Honest take

No red flags in our notes so far

• One of the strongest open-weight coding models, and Revolut's AI infrastructure lead says it is now the bank's default for about 2,000 engineers, self-hosted on roughly 56 B200 GPUs. That is the company's own account from a conference talk, not an audit.
• It is a Chinese lab, so many companies, especially regulated ones, will only use the weights on their own or a trusted provider's infrastructure, not Z.ai's China-based API. Revolut says it does exactly that.
• Self-hosting isn't plug-and-play: Revolut called the out-of-the-box performance terrible and spent months tuning before it hit its latency targets. Text-only, so images need another model.
• Pricing from docs.z.ai (3 Nov 2026). We haven't tested it hands-on.

Our Hype or Fact? checks

• Partly factRevolut moved ~2,000 engineers to self-hosted GLM at a tenth of the cost? 3 Nov

Pricing

API per million tokens: GLM-5.3 US$1.40 input, US$0.26 cached input, US$4.40 output; GLM-5.3-Flash US$0.15 / US$0.03 / US$0.50 (docs.z.ai pricing, checked 3 Nov 2026). The GLM Coding Plan starts at US$18/month (Lite, Pro and Max tiers, with 5-hour and weekly usage limits). Open weights are free to download and self-host under their licence.

Prices as listed by the maker when we last checked (2026-11-03); usually in US dollars. Check the website for current prices.

Latest update

• 3 November 2026: Revolut's David Abitbol said at AI Engineer NYC (13 Oct) and GTC Berlin (Oct) that Revolut has self-hosted GLM for its engineers since April 2026 (GLM-5.1, now GLM-5.3) on about 56 B200 GPUs, at a blended US$0.30 per million tokens. He also reported 70% better lead time. These are Revolut's own unaudited numbers; frontier models from Anthropic and OpenAI are kept for the hardest problems.
• 25 August 2026: GLM-5.3 weights released on Hugging Face: 753B-parameter MoE, 1M context, text-only, under the GLM-5.3 License (MIT-like, with a security-review clause for very large model-hosting companies). GLM-5.3-Flash is MIT.

Is GLM (Z.ai) worth it? Quick answers