AGI Agent Model Selection Guide
December 21, 2025 Β· View on GitHub
δΈζ | English
AGI Agent supports multiple AI models. This guide helps you choose the most suitable model based on your needs and budget.
π Recommended Models
Claude Sonnet 4 (β Highly Recommended)
Suitable for: Complex tasks requiring high accuracy and detailed responses
- β
Advantages:
- Extremely high intelligence and understanding capability
- Excellent code generation quality
- Detailed responses with deep analysis
- Outstanding tool calling capabilities
- β Disadvantages:
- Relatively high price
- Medium response speed
- Occasionally overly cautious behavior
- π° Price Level: $$$$
- π― Best Use Cases:
- Complex code architecture design
- Detailed technical analysis
- Advanced problem solving
- Multi-round complex tasks
Configuration Example:
python agia.py --model claude-3-5-sonnet-20241022 --api-key your_key -r "Your task"
OpenAI GPT-4 Turbo
Suitable for: Users needing fast and reliable performance
- β
Advantages:
- Fast response speed
- High accuracy
- Stable tool calling
- Complete ecosystem
- β Disadvantages:
- High price (but cheaper than Claude)
- Sometimes brief responses
- π° Price Level: $$$
- π― Best Use Cases:
- General development tasks
- Rapid iterative development
- Real-time interactive scenarios
- Balanced performance needs
Configuration Example:
python agia.py --model gpt-4-turbo --api-key your_key -r "Your task"
DeepSeek V3 (π° Best Value)
Suitable for: Users focusing on cost-effectiveness and accuracy
- β
Advantages:
- Extremely economical pricing
- Accurate code generation
- Fewer hallucination issues
- Clear thinking process
- β Disadvantages:
- Relatively concise output
- Lower creativity
- Average performance on some advanced tasks
- π° Price Level: $$
- π― Best Use Cases:
- Code optimization and refactoring
- Bug fixes
- Direct implementation tasks
- Budget-limited projects
Configuration Example:
python agia.py --model deepseek-chat --api-base https://api.deepseek.com --api-key your_key -r "Your task"
Kimi K2 (π Domestic Excellence)
Suitable for: Users needing Chinese optimization and long context
- β
Advantages:
- Strong Chinese understanding capability
- Ultra-long context support
- Reasonable pricing
- Optimized for Chinese development scenarios
- β Disadvantages:
- Relatively weak international support
- Average performance on some English tasks
- π° Price Level: $$$
- π― Best Use Cases:
- Chinese project development
- Large document processing
- Long conversation tasks
- Localization needs
Configuration Example:
python agia.py --model kimi --api-base https://api.moonshot.cn/v1 --api-key your_key -r "Your task"
Qwen2.5-7B-Instruct (π Free Trial)
Suitable for: Learning trials and simple tasks
- β
Advantages:
- Completely free to use
- Good Chinese support
- Basic task processing capability
- Quick response
- β Disadvantages:
- Limited intelligence level
- Average performance on complex tasks
- Weak tool calling capability
- π° Price Level: FREE
- π― Best Use Cases:
- Learning and experimentation
- Simple code generation
- Basic task processing
- Zero budget scenarios
Configuration Example:
python agia.py --model Qwen/Qwen2.5-7B-Instruct --api-base https://api.siliconflow.cn/v1 --api-key your_free_key -r "Your task"
π Model Comparison Table
| Model | Intelligence | Speed | Chinese Support | Cost | Best Use |
|---|---|---|---|---|---|
| Claude Sonnet 4 | βββββ | βββ | ββββ | π°π°π°π° | Complex projects |
| GPT-4 Turbo | ββββ | βββββ | βββ | π°π°π° | General development |
| DeepSeek V3 | ββββ | ββββ | ββββ | π°π° | Budget projects |
| Kimi K2 | ββββ | βββ | βββββ | π°π°π° | Chinese projects |
| Qwen2.5-7B | βββ | ββββ | ββββ | Free | Simple tasks |
π― Selection Recommendations
Choose by Project Type
π’ Enterprise Projects
Recommended: Claude Sonnet 4 or GPT-4 Turbo
- Need high-quality code generation
- Require detailed technical analysis
- Relatively sufficient budget
πΌ Commercial Projects
Recommended: DeepSeek V3 or Kimi K2
- Balance cost and performance
- Suitable for medium complexity tasks
- Excellent cost-effectiveness
π Learning and Experimentation
Recommended: Qwen2.5-7B or DeepSeek V3
- Limited budget or free
- Suitable for learning programming
- Simple task processing
π¨π³ Chinese Projects
Recommended: Kimi K2 or DeepSeek V3
- Excellent Chinese understanding
- Good localization support
- Fits domestic usage habits
Choose by Budget
High Budget (>$100/month)
- Claude Sonnet 4 - Highest quality
- GPT-4 Turbo - Speed and quality balance
Medium Budget ($20-100/month)
- DeepSeek V3 - Best cost-effectiveness
- Kimi K2 - First choice for Chinese projects
Low Budget/Free
- Qwen2.5-7B - Completely free
- DeepSeek V3 - Extremely low cost
βοΈ Configuration Guide
Configuration File Settings
Configure your chosen model in config/config.txt:
# Claude Sonnet 4
api_key=your_anthropic_key
api_base=https://api.anthropic.com
model=claude-3-5-sonnet-20241022
# GPT-4 Turbo
api_key=your_openai_key
api_base=https://api.openai.com/v1
model=gpt-4-turbo
# DeepSeek V3
api_key=your_deepseek_key
api_base=https://api.deepseek.com
model=deepseek-chat
# Kimi K2
api_key=your_kimi_key
api_base=https://api.moonshot.cn/v1
model=kimi
# Qwen2.5-7B (Free)
api_key=your_siliconflow_key
api_base=https://api.siliconflow.cn/v1
model=Qwen/Qwen2.5-7B-Instruct
Command Line Configuration
You can also specify models directly via command line:
# Temporarily use different models
python agia.py --model MODEL_NAME --api-key YOUR_KEY --api-base API_BASE -r "Task description"
π§ Optimization Recommendations
Performance Optimization
High-end Models (Claude/GPT-4)
truncation_length=15000
summary_trigger_length=120000
summary_max_length=8000
Budget Models (DeepSeek/Kimi)
truncation_length=10000
summary_trigger_length=80000
summary_max_length=5000
Free Models (Qwen)
truncation_length=6000
summary_trigger_length=50000
summary_max_length=3000
Tool Calling Optimization
Models supporting native tool calling:
- Claude Sonnet 4
- GPT-4 Turbo
- DeepSeek V3
Models requiring chat-based tool calling:
- Some local models
- Early version models
# Auto-detect or manual setting
Tool_calling_format=True # Recommended to keep default
Common Issues
1. Difficulty Choosing Models
Recommended Process:
- Clarify budget range
- Determine project complexity
- Consider language preference (Chinese/English)
- Start with recommended models for trial
2. API Configuration Issues
- Ensure API key is valid
- Check api_base address
- Verify model name is correct
3. Unsatisfactory Performance
- Try adjusting truncation parameters
- Check if task description is clear
- Consider upgrading to higher-end models
4. Cost Control
- Set reasonable truncation length
- Enable summary features
- Choose budget models
π Model Switching
AGI Agent supports switching models anytime without restarting tasks:
# Current task using DeepSeek
python agia.py --model deepseek-chat -r "Start task"
# Switch to Claude when needing higher quality
python agia.py --model claude-3-5-sonnet-20241022 -c # Continue previous task
Choosing the right model is key to successfully using AGI Agent. It's recommended to start with cost-effective models and adjust gradually based on actual needs.