AI Models

🎓 Claude 3 Opus

Model ID: anthropic/claude-3-opus

Context: 200,000 tokens

Strengths: Complex reasoning, large codebases, architecture design

Best for: Major refactoring, system design, complex debugging

Speed: ★★★☆☆ | Quality: ★★★★★

./abov3-linux-x64 run -m anthropic/claude-3-opus "Your complex task"

⚡ Claude 3 Sonnet

Model ID: anthropic/claude-3-sonnet

Context: 200,000 tokens

Strengths: Fast responses with high quality, great for iterative development

Best for: Daily coding tasks, code reviews, documentation

Speed: ★★★★☆ | Quality: ★★★★☆

./abov3-linux-x64 run -m anthropic/claude-3-sonnet "Your task"

🏃 Claude 3 Haiku

Model ID: anthropic/claude-3-haiku

Context: 200,000 tokens

Strengths: Lightning fast, cost-effective, good for simple tasks

Best for: Quick queries, simple refactoring, syntax questions

Speed: ★★★★★ | Quality: ★★★☆☆

./abov3-linux-x64 run -m anthropic/claude-3-haiku "Quick question"

🧠 GPT-4

Model ID: openai/gpt-4

Context: 128,000 tokens

Strengths: General knowledge, creative solutions, multi-modal

Best for: Creative coding, architectural decisions, complex algorithms

Speed: ★★★☆☆ | Quality: ★★★★★

./abov3-linux-x64 run -m openai/gpt-4 "Your task"

⚡ GPT-4 Turbo

Model ID: openai/gpt-4-turbo

Context: 128,000 tokens

Strengths: Faster than GPT-4, same quality

Best for: Balanced performance and speed needs

Speed: ★★★★☆ | Quality: ★★★★★

💨 GPT-3.5 Turbo

Model ID: openai/gpt-3.5-turbo

Context: 16,000 tokens

Strengths: Very fast, low cost, good for simple tasks

Best for: Quick answers, boilerplate code, simple scripts

Speed: ★★★★★ | Quality: ★★★☆☆

🔧 Qwen2.5-Coder

Model IDs:

• ollama/qwen2.5-coder:1.5b - Tiny, very fast
• ollama/qwen2.5-coder:7b - Balanced
• ollama/qwen2.5-coder:14b - Best quality
• ollama/qwen2.5-coder:32b - Professional

Strengths: Specialized for coding, excellent performance

Requirements: 4GB+ VRAM (7B), 8GB+ VRAM (14B)

# Install and use
ollama pull qwen2.5-coder:14b
./abov3-linux-x64 run -m ollama/qwen2.5-coder:14b "Generate code"

🦙 Llama 3

Model IDs:

• ollama/llama3:8b - Fast, general purpose
• ollama/llama3:70b - Powerful (needs 32GB+ RAM)

Strengths: Well-rounded, good reasoning

ollama pull llama3:8b
./abov3-linux-x64 run -m ollama/llama3:8b "Your task"

🌟 Mistral

Model IDs:

• ollama/mistral:7b - Fast and efficient
• ollama/mixtral:8x7b - MoE architecture, powerful

Strengths: Efficient, good for general tasks

ollama pull mistral:7b
./abov3-linux-x64 run -m ollama/mistral:7b "Your task"

💻 CodeLlama

Model IDs:

• ollama/codellama:7b - Code generation
• ollama/codellama:13b - Better quality
• ollama/codellama:34b - Best quality

Strengths: Specialized for code, supports many languages

Setting Default Model

# In config file
{
  "defaultModel": "anthropic/claude-3-sonnet"
}

# Via environment variable
export ABOV3_MODEL="ollama/qwen2.5-coder:14b"

# Via command line
./abov3-linux-x64 -m openai/gpt-4

Switching Models

# In TUI
/model anthropic/claude-3-opus

# For single command
./abov3-linux-x64 run -m ollama/mistral:7b "Quick task"

# List available models
./abov3-linux-x64 models

Model-Specific Settings

{
  "models": {
    "anthropic/claude-3-opus": {
      "temperature": 0.7,
      "maxTokens": 4096
    },
    "ollama/qwen2.5-coder:14b": {
      "temperature": 0.3,
      "topP": 0.9
    }
  }
}

🚀 Optimizing Model Usage

• Use faster models for simple tasks: Don't use Opus for syntax questions
• Local models for iteration: Use Ollama during development to save costs
• Streaming for feedback: Enable streaming for better UX
• Context management: Clear context when switching tasks
• Temperature tuning: Lower (0.3) for code, higher (0.7) for creative tasks

💰 Cost Optimization

• Development: Use local Ollama models
• Testing: Use Haiku or GPT-3.5 Turbo
• Production: Use Sonnet for balance
• Complex tasks: Reserve Opus for when needed
• OAuth: Use Anthropic OAuth if you have Claude Pro/Max