Model Guide

Recommendations for choosing OpenRouter models to use with Claude Code.

Quick Recommendations

Use Case	Model	Notes
Coding (Best)	qwen/qwen-2.5-coder-32b-instruct	Optimized for code
Fast Tasks	z-ai/glm-4.5-air	Very fast responses
General Purpose	anthropic/claude-3.5-sonnet	Versatile
Budget	anthropic/claude-3-haiku	Cheapest option

Default Configuration

The launcher uses these defaults if not specified:

ANTHROPIC_MODEL=qwen/qwen-2.5-coder-32b-instruct
ANTHROPIC_SMALL_FAST_MODEL=z-ai/glm-4.5-air

Models by Category

Coding Specialists

Best for writing, reviewing, and debugging code.

Model	Strengths
qwen/qwen-2.5-coder-32b-instruct	Excellent code generation, large context
deepseek/deepseek-coder	Strong coding, good at explanations
codellama/codellama-70b-instruct	Meta’s code-focused model

Fast/Budget Models

Best for quick operations, initial drafts, or cost-sensitive use.

Model	Strengths
z-ai/glm-4.5-air	Very fast, good quality
anthropic/claude-3-haiku	Fast, cheapest Claude model
google/gemini-flash-1.5	Fast, good for simple tasks

General Purpose

Good all-around models for varied tasks.

Model	Strengths
anthropic/claude-3.5-sonnet	Excellent reasoning, versatile
openai/gpt-4-turbo	Strong capabilities, well-known
google/gemini-pro-1.5	Large context, multimodal

Large Context

Best when working with many files or long documents.

Model	Context	Notes
anthropic/claude-3.5-sonnet	200K	High quality
google/gemini-pro-1.5	1M	Largest context
qwen/qwen-2.5-coder-32b-instruct	128K	Code-focused

Choosing a Model

For Development Work

Recommended: qwen/qwen-2.5-coder-32b-instruct

• Excellent at code generation and refactoring
• Good understanding of programming patterns
• Large context window for multi-file projects

For Quick Tasks

Recommended: z-ai/glm-4.5-air

• Very fast responses
• Good enough for simple edits
• Cost-effective for high-volume use

For Complex Analysis

Recommended: anthropic/claude-3.5-sonnet

• Strong reasoning capabilities
• Good at understanding nuanced requirements
• Excellent for code review

Model Pairing Strategy

Configure two models for different purposes:

# Main model for complex work
ANTHROPIC_MODEL=qwen/qwen-2.5-coder-32b-instruct

# Fast model for quick operations
ANTHROPIC_SMALL_FAST_MODEL=z-ai/glm-4.5-air

Claude Code uses:

• Main model for complex tasks, code generation, multi-file edits
• Small/fast model for quick completions, simple operations

Checking Available Models

Browse all models at: https://openrouter.ai/models

Filter by:

• Capability (chat, code, etc.)
• Context length
• Pricing
• Speed

Pricing Considerations

OpenRouter charges per token. Key factors:

1. Input tokens - Your prompts and context
2. Output tokens - Model responses
3. Model pricing - Varies significantly

Check pricing at: https://openrouter.ai/models

Cost Optimization Tips

1. Use fast/cheap models for simple tasks
2. Be concise in prompts
3. Use appropriate context (don’t send unnecessary files)
4. Monitor usage at https://openrouter.ai/activity

Testing a New Model

1. Update your .env:

   ANTHROPIC_MODEL=new-model-id
   

2. Restart the launcher

3. Test with a simple task first

4. Check response quality and speed

Model Compatibility Notes

Not all models support all features:

Feature	Requirement
Tool/function calling	Model must support functions
Streaming	Most models support this
System prompts	Most models support this
Long context	Check model’s context limit

The proxy handles format conversion, but the underlying model must support the feature.

See Also

• OpenRouter Models - Full model list
• Configuration Guide - Setting up models
• Architecture - How the proxy works