// Halo
Use GLM (Zhipu / Z.ai) without a subscription — GLM-5 on Halo
Zhipu's GLM family — a capable, efficient open model line — served peer-to-peer on Halo. No account, no subscription. Pay only per prompt.
GLM is Zhipu AI’s model family (also shipped as Z.ai) — a capable, cost-efficient line that’s popular for general chat, reasoning and agentic work. As with most Chinese models, the friction is access: accounts, regional availability, and questions about data handling.
Halo removes that. Independent operators serve GLM (alongside 140+ other models) and you reach them directly — no account, no subscription, no phone number.
No subscription — pay per prompt
Top up a wallet once and pay per prompt in USDC on Base. There’s no monthly plan and nothing to cancel; you only pay for what you use.
Reachable from anywhere
No central app to geo-block means you can use GLM wherever an operator serves it — no region lock, no “not available in your country.”
Private by default, verifiable by design
Your prompt is encrypted to the operator, and confidential mode hides it even from the operator inside a hardware enclave (TEE). Every response carries a statistical proof that the real GLM model produced it.
Ready to try it? Open the app and send a prompt, or read how to use Halo on the web. Prefer another Chinese model? DeepSeek and Qwen are on the network too.
Frequently asked questions
Is GLM free on Halo?
No — but there's no subscription and no monthly fee. You pay per prompt in USDC on Base, so you only pay for what you use. No account, no card, no phone number. And because independent operators compete for every request, prices are often among the lowest you'll find for that model anywhere.
Do I need a Zhipu / Z.ai account?
No. On Halo you connect a wallet and send a prompt — there's no signup with Zhipu or any provider, and your prompt isn't tied to a personal account.
Which GLM models can I use?
Operators on the network serve the GLM line, including GLM-4.7 and the GLM-5 family (with fast and turbo variants). Availability shifts as operators come online.
Is it private and trustworthy?
Your prompt is encrypted to the operator, and confidential mode can hide it even from the operator inside a hardware enclave (TEE). Every result carries a proof the real model produced it.