Overview
Zhipu AI (智谱) is a Beijing-based lab behind the GLM family of large language models. Its international brand is Z.ai, where you can chat directly at chat.z.ai, while developers integrate via an OpenAI-compatible API. The current lineup spans the flagship GLM-5.2 (released June 2026 with a 1M-token context) and the speed-tuned GLM-4.7 / GLM-4.5-Flash, which remain free for registered users. Unlike many Western labs, Zhipu releases open-weight variants under permissive licenses, so teams can self-host or fine-tune.
What makes GLM worth a slot here is the combination of breadth and price. The free Flash models cover chat, vision, web search, and agent mode with high daily quotas, while paid Coding Plans slot cleanly into Claude Code, Cline, and OpenCode. In our evaluation, GLM leads peers on bilingual Chinese–English quality and agentic coding benchmarks, and its inference is fast enough for interactive use. It is not the absolute best at pure English reasoning, but for the price it is a serious alternative to the big three.
Key Features
- Bilingual Chat & Reasoning — GLM-5.2 handles Chinese and English with strong reasoning, coding, and long-context retention up to 1M tokens.
- Free Flash Tier — GLM-4.7-Flash and GLM-4.5-Flash stay free for all registered users, ideal for prototyping and light daily use.
- OpenAI-Compatible API — Drop-in replacement for most OpenAI SDKs; supports tool calling for agent workflows.
- Open Weights — MIT-licensed models available on Hugging Face for self-hosting and commercial use.
- Agent & Coding Plans — Subscription tiers from ~$18/month plug into popular coding agents with heavier quotas.
Pricing
| Plan | Price | What’s included |
|---|---|---|
| Free | $0 | GLM-4.7-Flash / GLM-4.5-Flash, high daily limits, web search, agent mode |
| GLM-5.2 API | ~$1.40/M in, ~$4.40/M out | Flagship model, pay-as-you-go, OpenAI-compatible |
| Coding Lite | ~$18/mo | GLM-5.2 access, ~80 prompts per 5 hours |
| Coding Pro | ~$72/mo | Higher quota, ~400 prompts per 5 hours |
| Coding Max | ~$160/mo | ~1,600 prompts per 5 hours, priority throughput |
Comparison
vs. DeepSeek: Both are Chinese open-weight labs; DeepSeek leads on price predictability and Western community adoption, while GLM leads on bilingual quality and agentic coding benchmarks. vs. Qwen: Qwen has broader language support and a stronger generalist profile; GLM wins on cost-per-token and open-weight availability. vs. Claude: Claude leads on English prose and enterprise polish, but GLM is far cheaper and self-hostable, making it the better pick for budget-sensitive builders.