How to Use GLM-4.7 with Claude Code and Cursor (2026)

GLM-4.7 is Zhipu AI's latest large language model and has been gaining attention for its strong coding performance, competitive pricing, and generous API limits. If you are looking to diversify your AI coding tools beyond Claude and GPT, integrating GLM-4.7 into your existing workflow with Claude Code and Cursor is straightforward.

This guide walks you through setting up GLM-4.7 as an alternative or complementary model in both tools.

What Is GLM-4.7?

GLM-4.7 is the latest model in the GLM (General Language Model) series from Zhipu AI, a Beijing-based AI company. The model is competitive with Claude Sonnet and GPT-4o on coding benchmarks while being significantly cheaper.

GLM-4.7 Key Specs

Specification	GLM-4.7
Developer	Zhipu AI
Context window	128K tokens
Multilingual	English, Chinese, and 25+ languages
Coding performance	Competitive with Claude Sonnet 4
API pricing	~$0.50/M input, ~$1.50/M output
Open weights	Partially (some variants)
Tool calling	Yes
Vision	Yes (multimodal variant)

How GLM-4.7 Compares

Model	Coding Score*	Context	Input Cost ($/M)	Output Cost ($/M)
GLM-4.7	82.1	128K	~$0.50	~$1.50
Claude Sonnet 4	85.3	200K	$3.00	$15.00
GPT-4o	83.7	128K	$2.50	$10.00
DeepSeek V3	81.5	128K	$0.27	$1.10
Gemini 2.5 Pro	84.2	1M	$1.25	$10.00

*Approximate composite coding benchmark scores. Actual performance varies by task.

GLM-4.7 offers strong coding ability at a fraction of the cost of Claude or GPT-4o, making it an attractive option for high-volume coding tasks.

Getting a GLM-4.7 API Key

Option 1: Direct from Zhipu AI

Go to open.bigmodel.cn (Zhipu's developer platform).
Create an account and verify your identity.
Navigate to the API Keys section.
Generate a new API key.
Copy and store the key securely.

Zhipu typically offers free credits for new accounts.

Option 2: Via OpenRouter

OpenRouter provides access to GLM-4.7 through a unified API that uses the OpenAI-compatible format:

Go to openrouter.ai and create an account.
Add credits to your account.
Copy your OpenRouter API key.
The model ID for GLM-4.7 on OpenRouter is typically zhipu/glm-4.7.

OpenRouter is the easier option if you already use it for other models, since it provides a standardized OpenAI-compatible API.

Using GLM-4.7 with Cursor

Cursor supports custom model providers through its settings. Here is how to add GLM-4.7.

Method 1: Via OpenRouter (Recommended)

Open Cursor Settings (gear icon or Cmd+,).
Navigate to Models.
Under OpenAI API Key, enter your OpenRouter API key.
Set the Base URL to https://openrouter.ai/api/v1.
In the model list, click + Add Model and enter zhipu/glm-4.7.

You can now select GLM-4.7 from the model dropdown in Cursor's Composer or Chat.

Method 2: Via Zhipu's OpenAI-Compatible API

Zhipu AI provides an OpenAI-compatible endpoint. Configure Cursor to use it directly:

Open Cursor Settings > Models.
Set the Base URL to https://open.bigmodel.cn/api/paas/v4.
Enter your Zhipu API key.
Add the model name: glm-4.7.

Cursor Configuration Example

After setup, your Cursor model list should look something like this:

Model	Provider	Use Case
claude-sonnet-4	Anthropic	Primary coding model
glm-4.7	Zhipu/OpenRouter	Cost-effective alternative
gpt-4o	OpenAI	General tasks
deepseek-v3	DeepSeek	Budget option

You can switch between models per conversation. Use Claude Sonnet for complex architecture decisions and GLM-4.7 for routine coding tasks to save costs.

Using GLM-4.7 in Cursor Composer

Once configured, open Composer (Cmd+I or Ctrl+I) and select GLM-4.7 from the model dropdown:

Model: glm-4.7

Prompt: Refactor the UserService class to use the repository
pattern. Create a UserRepository interface and a
PostgresUserRepository implementation.

Cursor will use GLM-4.7 to plan and execute the multi-file edit, just as it would with any other model.

Using GLM-4.7 with Claude Code

Claude Code is Anthropic's CLI tool and primarily uses Claude models. However, you can configure it to use alternative models through custom API configurations.

Method 1: Using the --model Flag with Custom Provider

Claude Code supports the --model flag for specifying models. To use GLM-4.7, you need to configure a custom provider:

# Set environment variables for OpenRouter
export OPENROUTER_API_KEY="sk-or-xxxxxxxxxxxxx"

# Use Claude Code with a custom model via OpenRouter
claude --model openrouter/zhipu/glm-4.7

Method 2: Using the Model Configuration

Add GLM-4.7 to your Claude Code configuration. Edit ~/.claude/settings.json:

{
  "models": {
    "glm-4.7": {
      "provider": "openrouter",
      "apiKey": "sk-or-xxxxxxxxxxxxx",
      "model": "zhipu/glm-4.7"
    }
  }
}

Then switch models in an interactive session:

claude
> /model glm-4.7

Method 3: Hybrid Workflow

The most practical approach is using Claude as your primary model and routing specific tasks to GLM-4.7 when appropriate:

# Use Claude for complex architecture decisions
claude "design the database schema for a multi-tenant SaaS application"

# Use GLM-4.7 for routine CRUD operations (via OpenRouter/custom setup)
claude --model glm-4.7 "generate the CRUD endpoints for the User model"

Practical Integration Patterns

Pattern 1: Cost-Optimized Workflow

Use GLM-4.7 for high-volume, routine tasks and save Claude for complex work:

Task Type	Recommended Model	Why
Boilerplate generation	GLM-4.7	Cheap, fast, good enough
CRUD endpoints	GLM-4.7	Routine patterns
Unit test generation	GLM-4.7	Pattern-based, volume work
Architecture design	Claude Sonnet/Opus	Needs deep reasoning
Complex debugging	Claude Sonnet/Opus	Needs broad context
Code review	GLM-4.7 or Claude	Either works well

Pattern 2: Second-Opinion Workflow

Use GLM-4.7 as a second opinion on Claude's output:

# Get Claude's solution
claude "implement rate limiting middleware for Express.js"

# Compare with GLM-4.7's approach
claude --model glm-4.7 "implement rate limiting middleware for Express.js"

Different models have different strengths. Comparing outputs can help you find the best implementation.

Pattern 3: Language-Specific Optimization

GLM-4.7 has particularly strong performance on:

Python and JavaScript/TypeScript code
Chinese-language documentation and comments
Data processing and algorithm tasks
Test generation

# GLM-4.7 excels at generating well-documented Python code
claude --model glm-4.7 "create a Python data pipeline that reads CSV files,
cleans the data, and outputs to Parquet format. Add type hints and docstrings."

Benchmarking GLM-4.7 for Your Use Case

Before committing to GLM-4.7 for production workflows, benchmark it on your specific tasks. Here is a simple approach:

# Create a test prompt file
cat << 'EOF' > test-prompt.txt
Create a TypeScript function that:
1. Accepts an array of user objects with name, email, and age
2. Validates all fields (email format, age > 0)
3. Removes duplicates by email
4. Sorts by age ascending
5. Returns the cleaned array
Include proper TypeScript types and error handling.
EOF

# Test with Claude
claude -p < test-prompt.txt > claude-output.ts

# Test with GLM-4.7
claude -p --model glm-4.7 < test-prompt.txt > glm-output.ts

# Compare the outputs
diff claude-output.ts glm-output.ts

Evaluate the outputs on:

Criteria	Weight
Correctness	High
Code style and readability	Medium
Type safety	Medium
Edge case handling	High
Documentation/comments	Low
Response time	Low

Troubleshooting

"Model not found" error in Cursor: Make sure you have the correct model ID. On OpenRouter, it is typically zhipu/glm-4.7. Check the OpenRouter models page for the exact ID.

Authentication failures: Verify your API key is correct and has sufficient credits. Zhipu keys start with a specific prefix. OpenRouter keys start with sk-or-.

Slow response times: GLM-4.7 servers are primarily hosted in China. If you are outside Asia, response latency may be higher than Claude or GPT. Using OpenRouter as an intermediary can sometimes improve routing.

Chinese characters in code comments: GLM-4.7 may occasionally generate Chinese comments or variable names, especially for ambiguous prompts. Add explicit instructions like "Write all comments and documentation in English" to your prompt or custom instructions.

Rate limit errors: Zhipu's free tier has rate limits. Upgrade to a paid tier or use OpenRouter for more generous limits.

Wrapping Up

GLM-4.7 is a strong addition to any AI coding workflow, especially when used strategically alongside Claude and GPT. Its competitive coding performance at a fraction of the cost makes it ideal for high-volume routine tasks, while you reserve premium models for complex reasoning and architecture work. Both Cursor and Claude Code can integrate with GLM-4.7 through OpenRouter or direct API configuration.

If your projects involve AI-generated media like images, video, or talking avatars, check out Hypereal AI for a unified API that handles all media generation with pay-as-you-go pricing.

Try Hypereal AI free -- 35 credits, no credit card required.