Z.ai (Zhipu AI) - Complete Knowledge Base

Last Updated: February 21, 2026 @ 16:30 EST Provider: Zhipu AI (Z.ai / BigModel.cn) Purpose: Comprehensive brain dump of all Z.ai knowledge for integration and reference

Related:

• Z.ai Coding Subscription - Quick tier reference and pricing
• Model Routing - Configuring complexity-aware routing
• Configuration Guide - API key setup and baseUrl
• Model Concurrency Findings - Real-world concurrency behavior

See Also: Z.ai Coding Subscription Quick Reference

Table of Contents

1. Executive Summary
2. Company Overview
3. Product Ecosystem
4. Subscription Tiers Deep Dive
5. Model Specifications
6. API Reference
7. Rate Limits & Quotas
8. Integrations
9. Competitive Analysis
10. Known Issues & Workarounds
11. Historical Changes
12. Community Tools
13. Cross-References

Executive Summary

What is Z.ai?

Z.ai (Zhipu AI / 智谱AI) is a Chinese AI company offering GLM (General Language Model) series models through a subscription-based coding plan. Their value proposition is “1/7 the cost of Claude, 3x the usage”.

Key Metrics

Why Z.ai Matters

• Cost Effective: Significantly cheaper than Anthropic Claude for coding tasks
• High Quota: 5-hour rolling window with generous prompt allowances
• Compatible: Works with Claude Code, Cursor, Cline, Roo Code, and more
• OpenAI/Anthropic Compatible: Easy drop-in replacement
• Growing Rapidly: 2.7M paid users as of December 2025

Company Overview

Zhipu AI (智谱AI)

Platform URLs

Product Ecosystem

1. GLM Coding Plan (Subscription)

Target: Individual developers using AI coding assistants

Features:

• Subscription-based pricing (not token-based)
• 5-hour rolling window quotas
• Compatible with 10+ coding tools
• MCP tools integration (vision, web search, web reader)
• Special API endpoints (/api/coding/paas/v4)

Tiers: Lite, Pro, Max

2. Standard API (Pay-as-you-go)

Target: Developers building applications

Features:

• Token-based pricing
• Standard OpenAI-compatible endpoints
• Flexible usage without quotas
• Higher per-token cost than coding plan

Pricing Example:

• GLM-4.7: $0.60/1M tokens (input), $2.20/1M tokens (output)
• GLM-4.7-Flash: Free tier available (1 concurrent request limit)

3. Enterprise Solutions

Launched: October 2025

Features:

• Custom deployments
• Dedicated support
• SLA guarantees
• Volume pricing

4. Z Code IDE

Description: Lightweight AI IDE by Zhipu AI

Features:

• Multi-agent unified scheduling (Claude Code, Codex, Gemini)
• Dynamic thinking mechanism
• Security control system
• Version management
• Integrated toolchain

5. GLM-4.7-Flash (Free Tier)

Limits:

• 1 concurrent request
• Free for development/testing
• Production use requires paid tier

Subscription Tiers Deep Dive

Complete Pricing Table (Post February 12, 2026)

International Market (USD)

China Market (CNY)

Tier Feature Comparison

Value Proposition Analysis

Compared to Claude Pro ($20/month)

Compared to Claude Max (estimated)

Model Specifications

GLM Model Family

Model Capabilities

GLM-4.7 (Current Flagship)

Strengths:

• SOTA open-source coding model
• 128K-200K context window
• Native 55+ tokens/second generation speed
• Function calling support
• JSON output mode
• Thinking mode for complex reasoning

Best For:

• Large codebase analysis
• Complex refactoring tasks
• Multi-file code generation
• Architecture-level decisions

GLM-5 (New Release)

Improvements over GLM-4.7:

• 754B parameters (2x+ GLM-4.7)
• Approaches Claude Opus 4.5 performance
• Better reasoning capabilities
• Improved code understanding

Trade-offs:

• Significantly higher token consumption
• Pro/Max only (Lite not supported yet)
• Higher latency than GLM-4.7

GLM-4.5-Air (Cost Optimized)

Strengths:

• Lowest token cost
• Fast response times
• Good for simple tasks

Best For:

• Simple code completion
• Bug fixes in small files
• Quick queries
• High-volume, low-complexity tasks

Context Window Evolution

Context Equivalents:

• 128K ≈ 150 pages of complex documents
• 128K ≈ 200 pages of PPT
• 128K ≈ 1 hour of video
• 200K ≈ 300+ pages of code

API Reference

Base URLs

Authentication

Bearer Token (API Key)

curl -X POST "https://api.z.ai/api/paas/v4/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "glm-4.7-flash",
    "messages": [{"role": "user", "content": "Hello"}],
    "max_tokens": 4096
  }'

Anthropic-Compatible Format

curl -X POST "https://open.bigmodel.cn/api/anthropic/v1/messages" \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "glm-4.7",
    "max_tokens": 4096,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

OpenAI SDK Example

from openai import OpenAI

client = OpenAI(
    api_key="your-zai-api-key",
    base_url="https://api.z.ai/api/paas/v4/",
)

response = client.chat.completions.create(
    model="glm-4.7-flash",
    messages=[
        {"role": "user", "content": "Write a Python function to parse CSV"}
    ],
    max_tokens=4096,
)

print(response.choices[0].message.content)

Anthropic SDK Example

import anthropic

client = anthropic.Anthropic(
    api_key="your-zai-api-key",
    base_url="https://open.bigmodel.cn/api/anthropic"
)

message = client.messages.create(
    model="glm-4.7",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Write a Python function"}
    ]
)

print(message.content[0].text)

Environment Variables

For Claude Code

macOS/Linux:

export ANTHROPIC_BASE_URL="https://open.bigmodel.cn/api/anthropic"
export ANTHROPIC_API_KEY="your-zai-api-key"

Windows PowerShell:

$env:ANTHROPIC_BASE_URL = "https://open.bigmodel.cn/api/anthropic"
$env:ANTHROPIC_API_KEY = "your-zai-api-key"

Windows CMD:

set ANTHROPIC_BASE_URL=https://open.bigmodel.cn/api/anthropic
set ANTHROPIC_API_KEY=your-zai-api-key

For OpenAI-compatible tools

export OPENAI_BASE_URL="https://api.z.ai/api/paas/v4/"
export OPENAI_API_KEY="your-zai-api-key"

Available Models (for API calls)

Rate Limits & Quotas

Understanding the 5-Hour Rolling Window

How it works:

1. The system tracks all prompt usage within a sliding 5-hour window
2. Quota used 5+ hours ago automatically releases
3. No fixed reset times (unlike “daily” or “monthly” quotas)
4. Dynamic recovery based on consumption patterns

Example:

• At 10:00: Use 50 prompts (70 remaining)
• At 12:00: Use 30 prompts (40 remaining)
• At 14:00: The 50 prompts from 10:00 “expire” from the window (90 available)
• At 15:00: The 30 prompts from 12:00 “expire” (120 available, full refresh)

Quota Tables

5-Hour Window Limits

Estimated Monthly Limits

Note: These are estimates based on continuous usage. Actual limits depend on usage patterns.

MCP Tool Limits (Monthly)

Undocumented/Unknown Limits

The following information is NOT publicly documented:

Peak Hour Issues (Known)

Capacity Measures (Since January 23, 2026)

• Daily sales limited to 20% of capacity
• Daily quota refreshes at 10:00
• Auto-renewal subscribers unaffected by limits
• Active crackdown on malicious traffic and violations

Integrations

Supported Tools

Z.ai GLM Coding Plan is compatible with 10+ mainstream coding tools:

Claude Code Integration

Step 1: Set Environment Variables

# Mac/Linux
export ANTHROPIC_BASE_URL="https://open.bigmodel.cn/api/anthropic"
export ANTHROPIC_API_KEY="your-zai-coding-plan-api-key"

# Windows PowerShell
$env:ANTHROPIC_BASE_URL = "https://open.bigmodel.cn/api/anthropic"
$env:ANTHROPIC_API_KEY = "your-zai-coding-plan-api-key"

Step 2: Configure Model

In .claude/settings.json:

{
  "provider": "anthropic",
  "model": "glm-4.7"
}

Available Models for Claude Code:

• glm-4.7 (recommended)
• glm-4.6
• glm-4.5
• glm-4.5-air
• glm-5 (Pro/Max only)

Cursor Integration

Settings → API Providers → Custom

{
  "baseUrl": "https://api.z.ai/api/paas/v4/",
  "apiKey": "your-api-key",
  "model": "glm-4.7"
}

Cline Integration (VS Code)

Settings JSON:

{
  "cline.apiUrl": "https://api.z.ai/api/paas/v4/",
  "cline.apiKey": "your-api-key",
  "cline.model": "glm-4.7"
}

Roo Code Integration

Configuration:

{
  "provider": "custom",
  "baseUrl": "https://api.z.ai/api/paas/v4/",
  "apiKey": "your-api-key",
  "model": "glm-4.7"
}

Competitive Analysis

Chinese Market Comparison

*First-month promotional pricing

International vs US Providers

Performance Comparison (SWE Benchmark)

Known Issues & Workarounds

Issue #1: Undocumented Concurrency Limit

Description:

• GLM 4.7 appears to only allow 1 concurrent in-flight request, even on paid tiers
• This limit is not documented officially

Reported In:

• GitHub issue: anomalyco/opencode#8618

Workaround:

// Implement queue-based request handling
class GLMRequestQueue {
  constructor() {
    this.queue = [];
    this.processing = false;
  }

  async add(request) {
    return new Promise((resolve, reject) => {
      this.queue.push({ request, resolve, reject });
      this.process();
    });
  }

  async process() {
    if (this.processing || this.queue.length === 0) return;

    this.processing = true;
    const { request, resolve, reject } = this.queue.shift();

    try {
      const result = await fetch(request);
      resolve(result);
    } catch (error) {
      reject(error);
    } finally {
      this.processing = false;
      this.process(); // Process next
    }
  }
}

Issue #2: Peak Hour Slowdowns

Description:

• Weekdays 15:00-18:00 (3-6 PM) experience significant slowdowns
• Concurrent rate limiting errors
• Slower model responses

Workarounds:

1. Schedule heavy work for off-peak hours (before 15:00, after 18:00)
2. Upgrade to Max tier - guaranteed “no lag” during peak
3. Enable auto-renewal - more stable resource allocation
4. Use retry logic with exponential backoff

async function callGLMWithRetry(request, maxRetries = 3) {
  for (let i = 0; i < maxRetries; i++) {
    try {
      return await fetch(request);
    } catch (error) {
      if (error.status === 429 && i < maxRetries - 1) {
        const delay = Math.pow(2, i) * 1000; // 1s, 2s, 4s
        await new Promise(r => setTimeout(r, delay));
        continue;
      }
      throw error;
    }
  }
}

Issue #3: GLM-5 High Token Consumption

Description:

• GLM-5 uses significantly more quota per prompt than GLM-4.7
• Exact multiplier not documented
• Can exhaust quota faster than expected

Workaround:

• Use GLM-4.7 for routine tasks
• Reserve GLM-5 for complex reasoning only
• Monitor quota usage in real-time

const TASK_COMPLEXITY = {
  simple: { model: 'glm-4.5-air', threshold: 0.3 },
  medium: { model: 'glm-4.7', threshold: 0.7 },
  complex: { model: 'glm-5', threshold: 1.0 }
};

function selectModelForTask(complexity) {
  if (complexity < 0.3) return 'glm-4.5-air';
  if (complexity < 0.7) return 'glm-4.7';
  return 'glm-5';
}

Issue #4: Daily Purchase Limits

Description:

• Since January 23, 2026, daily sales limited to 20% of capacity
• New subscriptions may not be available every day

Workarounds:

1. Try early in the day - quota refreshes at 10:00
2. Enable auto-renewal - unaffected by daily limits
3. Purchase quarterly/yearly - lock in longer term

Common Error Messages

Historical Changes

February 12, 2026 - Price Increase

Impact: 30%+ price increase across all tiers

January 23, 2026 - Capacity Constraints

December 2025 - User Milestone

• 2.7 million paid users reached
• Enterprise version launched (October 2025)
• GLM-5 announced

October 2025 - Enterprise Launch

• Dedicated deployments
• Custom SLAs
• Volume pricing
• Priority support

Community Tools

OpenCode GLM Quota Monitor

Repository: github.com/guyinwonder168/opencode-glm-quota

Features:

• Real-time quota monitoring
• 5-hour token cycle tracking
• Monthly MCP usage statistics
• 24-hour rolling window tracking
• Model and MCP tool usage (web_search, web_reader)

Official SDKs

Third-Party Providers

Cross-References

Internal Project Documentation

Configuration Files

External Documentation

Key Management URLs

Quick Reference Card

Environment Variables

# Anthropic-compatible (recommended for Claude Code)
export ANTHROPIC_BASE_URL="https://open.bigmodel.cn/api/anthropic"
export ANTHROPIC_API_KEY="your-key-here"

# OpenAI-compatible
export OPENAI_BASE_URL="https://api.z.ai/api/paas/v4/"
export OPENAI_API_KEY="your-key-here"

Model Selection Guide

Tier Selection Guide

This knowledge base is maintained as a living document. Last comprehensive research: February 21, 2026.

Sources Used

This document aggregates information from:

• Official Z.ai documentation (docs.z.ai)
• Official BigModel documentation (docs.bigmodel.cn)
• GitHub issue reports
• CSDN technical blogs
• Chinese tech news sources (Sohu, QQ News, 51CTO, Sina)
• Third-party provider documentation (SophNet, UCloud, Qiniu, PPIO)
• Community tools and repositories
• User reports and testimonials

For the most current information, always check official sources at z.ai or open.bigmodel.cn.

Expanded Competitive Analysis (2026 Research)

Chinese AI Coding Assistants - Detailed Comparison

Alibaba Qwen (通义千问)

Strengths:

• Largest context window (1M tokens expandable)
• 370+ programming language support
• “Agentic training” with 800K verified coding tasks
• Outperforms GPT-4.1 on SWE-Bench

ByteDance (Doubao 豆包)

Strengths:

• Figma-to-code conversion
• Context engineering for requirements
• Multi-modal interaction
• Agent capabilities for autonomous delivery

Market Comparison Table

Key Finding: Chinese models are 5-53x cheaper than Western competitors while maintaining competitive performance.

Step-by-Step Integration Guides

Cursor IDE - Complete Setup

Prerequisites

• Cursor IDE installed
• Z.ai GLM Coding Plan subscription
• API Key (format: sk-sp-xxxxx for Coding Plan)

Step 1: Get Your API Key

1. Visit https://z.ai or https://www.bigmodel.cn/glm-coding
2. Go to Personal Center → API Keys
3. Create a new API Key
4. Important: Use the Coding Plan专属 API Key (format: sk-sp-xxxxx)
5. Do NOT use the general API Key (format: sk-xxxxx)

Step 2: Configure Cursor

1. Open Cursor IDE
2. Go to Settings → Models (or press Ctrl+,)
3. Under API Providers, select Custom or OpenAI Compatible
4. Configure as follows:

OpenAI API Key: [Your sk-sp-xxxxx key]
Base URL: https://open.bigmodel.cn/api/anthropic
Model: glm-4.7 (or glm-5 for Pro/Max subscribers)

Step 3: Verify Configuration

Test with a simple prompt:

Can you write a Python function to calculate fibonacci numbers?

If successful, you should see GLM-4.7 responding.

Troubleshooting

Cline (VS Code Extension) - Complete Setup

Prerequisites

• VS Code installed
• GLM Coding Plan subscription
• API Key from Zhipu AI console

Step 1: Install Cline

1. Open VS Code
2. Press Ctrl+Shift+X to open Extensions
3. Search for “cline”
4. Click Install

Step 2: Configure Cline

1. Click the Cline icon in the sidebar (or press Ctrl+Shift+A)
2. Click the Settings (gear icon)
3. Select “Use your own API Key”
4. Configure as follows:

Step 3: Advanced Configuration

For optimal performance with Cline:

{
  "apiProvider": "openai",
  "baseUrl": "https://open.bigmodel.cn/api/coding/paas/v4",
  "apiKey": "your-key-here",
  "modelId": "glm-4.7",
  "contextLength": 204800,
  "temperature": 0.7,
  "maxTokens": 8192
}

Roo Code - Complete Setup

Prerequisites

• Roo Code VS Code extension installed
• Z.ai GLM Coding Plan subscription

Step 1: Configure Roo Code

1. Open VS Code with Roo Code installed
2. Open Settings → search for “Roo Code”
3. Configure API settings:

{
  "rooCode.apiProvider": "openai",
  "rooCode.apiKey": "your-sk-sp-xxxxx-key",
  "rooCode.baseUrl": "https://api.z.ai/api/paas/v4/",
  "rooCode.model": "glm-4.7"
}

Step 2: Alternative Method (Environment Variables)

# Set environment variables
export OPENAI_API_KEY="your-sk-sp-xxxxx-key"
export OPENAI_BASE_URL="https://api.z.ai/api/paas/v4/"

# Restart VS Code for changes to take effect

Claude Code - Automated Setup

Using the Setup Helper

Z.ai provides an automated configuration tool:

npx @z_ai/coding-helper

Follow the prompts:

1. Select Coding Plan: China version (中国版)
2. Enter your API Key (sk-sp-xxxxx)
3. Select Coding Tool: Claude Code / Roo Code / Cursor / Cline
4. Select MCP Config: Select all or specific tools

Manual Configuration

For Windows (C:\Users\<username>\.claude\settings.json):

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "your_zhipu_api_key",
    "ANTHROPIC_BASE_URL": "https://open.bigmodel.cn/api/anthropic",
    "API_TIMEOUT_MS": "3000000",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1,
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-4.7",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-4.5-air"
  }
}

For macOS/Linux (~/.claude/settings.json):

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://open.bigmodel.cn/api/anthropic",
    "ANTHROPIC_API_KEY": "your_zhipu_api_key"
  }
}

Data Privacy & Security

Privacy Policy Information

Z.ai (International Platform)

Key Statement from Z.ai:

“We do not store content users provide or generate while using our services, including text prompts, images, or other input data.”

BigModel.cn (China Platform)

Data Types Collected:

• Essential Information: Required for basic functionality
• Optional Information: For enhanced features

User Rights:

• Access and manage personal information
• Delete account and data
• Opt-out of optional data collection
• Minor protection measures

Data Retention (Unknown/Undocumented)

Recommendations for Privacy-Conscious Users

1. Review Privacy Policy Directly: https://bigmodel.cn/dev/howuse/privacypolicy
2. Contact Enterprise Sales: For formal Data Processing Addendum
3. Check Data Residency: China-based vs Singapore-based options
4. Consider Local Deployment: For sensitive data (CodeGeeX offers this)

Reliability & SLA Information

SLA Status

Known Reliability Issues

Peak Hour Performance

API Error Handling

The Z.ai Python SDK documents these error types:

Incident History (2025-2026)

Note: Unlike OpenAI and Google Gemini (which had major outages in Dec 2025/Jan 2026), no specific Zhipu AI service-wide outages were documented in the search results.

Reliability Best Practices

1. Implement Retry Logic: Use exponential backoff for 429/500 errors
2. Monitor Quota: Track 5-hour window usage to avoid exhaustion
3. Off-Peak Scheduling: Schedule heavy work outside 15:00-18:00
4. Max Tier for Production: Consider upgrading for priority access
5. Fallback Models: Implement model fallback chains

Daily Purchase & Availability Guide

Current Availability Status (2026)

Purchase Strategy

Best Times to Purchase

1. Early morning (10:00 AM sharp) - Daily quota refresh
2. Mid-week (Tuesday-Thursday) - Lower competition
3. Avoid: Weekends, month-start, Monday mornings

Securing Your Subscription

“Sold Out” (售罄) Information

When subscriptions show as “sold out”:

• Wait until 10:00 AM next day for quota refresh
• Check availability throughout the day - cancellations may open spots
• Consider auto-renewal - guaranteed renewal

Automated Setup Tool

Using @z_ai/coding-helper

Z.ai provides an official setup helper:

npx @z_ai/coding-helper

Interactive Prompts

What the Tool Configures

• Environment variables
• API endpoints
• Model mappings
• MCP tool connections
• Context window settings

Updated Quick Reference Card

All Endpoints Summary

API Key Formats

Model Mapping by Task

Last comprehensive update: February 21, 2026 @ 16:30 EST

Sources (Expanded List)

This document aggregates information from:

• Official Z.ai documentation (docs.z.ai)
• Official BigModel documentation (docs.bigmodel.cn)
• GitHub issue reports and community discussions
• CSDN technical blogs and tutorials
• Chinese tech news (Sohu, QQ News, 51CTO, Sina)
• Third-party provider documentation (SophNet, UCloud, Qiniu, PPIO)
• Community tools and repositories
• User reports and testimonials
• Hacker News discussions on Z.ai services
• Industry analysis on Chinese AI coding assistants

Specific Sources Referenced

• Cline + GLM Tutorial: CSDN Article (Dec 22, 2025)
• Cline API Configuration: CSDN Guide (Dec 11, 2025)
• Claude Code + GLM Setup: CSDN Tutorial (Feb 15, 2026)
• Chinese AI Comparison: Multiple sources on 通义灵码, 豆包编程, CodeGeeX
• DeepSeek V3 Pricing: Industry comparison articles (2026)
• Privacy Policy: https://bigmodel.cn/dev/howuse/privacypolicy