Zhipu AI GLM
Listed at https://open.bigmodel.cn
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | GLM-5.3 (Workers AI): $1.40/M + $0.26/M cached; GLM-5.2: ¥8/M; GLM-5.3-Flash (Workers AI): $0.15/M; GLM-5: ¥4-6/M; GLM-4.7: ¥2-4/M; GLM-4.5-Air: ¥0.8/M; GLM-4.7-Flash: ¥0/M (free) | per million tokens |
| Output | GLM-5.3 (Workers AI): $4.40/M; GLM-5.2: ¥28/M; GLM-5.3-Flash (Workers AI): $0.50/M; GLM-5: ¥18-22/M; GLM-4.7: ¥8-16/M; GLM-4.5-Air: ¥2-8/M; GLM-4.7-Flash: ¥0/M (free) | per million tokens |
🤖 Supported Models (14)
✨ Pros
- ✓GLM-5.3 — strongest open-weights coding model (Aug 28, 2026 on Cloudflare Workers AI, same price as GLM-5.2, 6.1x Terminal Bench 3.0)
- ✓GLM-4-Flash completely free, ideal for high-volume light calls
- ✓CogView-3 image generation + CogVideoX video generation
- ✓Diverse model types (text+image+video)
- ✓GLM-4-Plus performance approaching GPT-4 level
- ✓Direct China access, stable and reliable
- ✓Well-known open-source (ChatGLM) community presence
⚠️ Cons
- ×GLM-4-Plus extremely expensive (¥50/M tokens)
- ×English capabilities behind DeepSeek/Qwen
- ×Multimodal model API documentation immature
- ×Free Flash model has usage limits
- ×Low international brand recognition
🎯 Best For
China-based app development; GLM-4-Flash free tier for low-cost prototyping; multimodal needs (text+image+video)
💰 Pricing & Plans
| Model | Context | Input (¥/M) | Output (¥/M) | Notes |
|---|---|---|---|---|
| GLM-5.2 (new) | 1M | ¥8 (¥2 cache-hit) | ¥28 | Long-task flagship, full dev-to-deploy |
| GLM-5.1 | up to 1M | ¥6–¥8 | ¥24–¥28 | Frontier reasoning, 32K-input tiering |
| GLM-5-Turbo | up to 1M | ¥5–¥7 | ¥22–¥26 | Balanced speed/cost for production |
| GLM-5 | up to 1M | ¥4–¥6 | ¥18–¥22 | Strong general workhorse |
| GLM-4.7 | 200K | ¥2–¥4 | ¥8–¥16 | Cost-efficient mid-tier |
| GLM-4.5-Air | 128K | ¥0.8 | ¥2–¥8 | Cheap high-volume option |
| GLM-4.7-FlashX | 200K | ¥0.5 | ¥3 | Very low-cost 200K context |
| GLM-4.7-Flash | 200K | Free | Free | Fully free, generous limits |
| GLM-4.6V | 128K | ¥1–¥2 | ¥3–¥6 | Vision (image/video) understanding |
🔧 API & Developer Experience
- •API Style: OpenAI-compatible chat/completions REST endpoints on open.bigmodel.cn with SSE streaming; most OpenAI SDKs and wrappers connect after a base-url and key swap.
- •SDKs & Ecosystem: First-party Python and Node SDKs plus an OpenAI-compatible client; Zhipu also publishes open weights in the ChatGLM/GLM lineage, so there is a large self-hosted community and tooling ecosystem.
- •Tool Calling & Agents: Native function/tool calling with structured JSON on the GLM-5.x line; GLM models are widely used as the default 'tool-use' backbone in domestic agent frameworks.
- •AutoGLM (Agent): AutoGLM / GLM-Phone is Zhipu's flagship agent that operates a phone/browser via screenshots and taps — a distinct differentiator for real-world task automation beyond pure text APIs.
- •Prompt Caching: Automatic prefix caching across the GLM-5.x and GLM-4.7 line — cache hits are billed at roughly a quarter of the input price (e.g. GLM-5.2 ¥2/M hit vs ¥8/M miss), cutting repeated-context cost sharply.
- •Vision & Real-time: GLM-4.6V / GLM-5V handle image and video input; GLM-4-Voice, CogTTS and GLM-ASR cover real-time speech — text, vision, audio and video under one console.
- •Billing & Region: Billed in CNY with prepaid recharge; a free tier (GLM-4.7-Flash) plus new-user token coupons make prototyping nearly zero-cost. Direct mainland access with no proxy required.
🎯 Agent & Long-Context (GLM-5.2 / AutoGLM)
Zhipu's defining strength is agentic, long-horizon AI built on a genuinely usable 1M-token context. GLM-5.2, the current flagship, holds project-grade engineering context in a single window and can carry a task from requirements through multi-platform deployment without reloading — with automatic prefix caching (¥2/M on cache-hit vs ¥8/M miss) that makes repository-scale reasoning economical. The GLM-5.x line tiers by input length (short-input pricing below 32K tokens), so interactive workloads stay cheap while long-context agents pay a fair premium. Zhipu's other pillar is AutoGLM (GLM-Phone), an autonomous agent that operates a phone or browser through screenshots and taps. Combined with an open-weight ChatGLM lineage and a domestic-first product, Zhipu is one of the strongest all-round domestic options for agent pipelines and whole-repository coding.
🌐 China Access & Latency
Zhipu AI is a Beijing-based company, and the open.bigmodel.cn API is native-first for mainland China: directly reachable without a proxy, no GFW circumvention, and no cross-border egress fees. Domestic latency from mainland data centers is typically low single-digit to low-teens milliseconds RTT to Zhipu's China-region endpoints, and billing is in CNY — removing the currency and invoicing friction carried by overseas providers. This makes Zhipu one of the easiest domestic options for China-based teams: register with a mainland phone number, recharge in RMB, and call a domestic endpoint. The main caveats are the usual China-cloud concerns — SLA and redundancy tie to Zhipu's mainland infrastructure, international documentation is secondary to the domestic focus, and frontier GLM-5.x models ship primarily through Zhipu's own platform rather than a broad Western reseller network.