Model Selection

Choose the perfect AI model for your needs

The Right Model for Every Task

Siya supports a comprehensive range of cutting-edge AI models, from powerful cloud-based options to privacy-focused local models. Understanding each model’s strengths helps you choose the perfect one for your specific needs, balancing performance, cost, and capabilities.

Model Overview

Cloud Models

Cutting-Edge AI at Your Fingertips

Access the most powerful AI models through secure API connections

Claude Models

Anthropic's Claude Family

Industry-leading models known for reasoning, safety, and helpful responses

Claude Opus 4

The Flagship Model

Claude Opus 4 represents the pinnacle of AI capability, excelling at:

• Complex reasoning and analysis
• Architecture design and planning
• Code generation and debugging
• Creative problem solving
• Multi-step task execution

Specifications

• Context: 200K tokens
• Strengths: Deep reasoning
• Speed: 20-30 tokens/sec
• Best for: Complex tasks
• Cost: Premium tier

Choose Opus 4 when accuracy and reasoning depth matter more than speed

Claude Sonnet 4

The Balanced Performer

Sonnet 4 offers an optimal balance of capability and efficiency:

• Fast response times
• Excellent code understanding
• Strong analytical skills
• Cost-effective for most tasks
• Great for iterative development

Specifications

• Context: 200K tokens
• Strengths: Balance
• Speed: 40-50 tokens/sec
• Best for: Daily coding
• Cost: Standard tier

Sonnet 4 is recommended for 80% of development tasks

OpenAI Models

OpenAI GPT Family

Versatile models with broad capabilities and extensive training

GPT-5

Next-Generation Intelligence

GPT-5 brings unprecedented capabilities:

• Multimodal understanding
• Advanced code generation
• System design expertise
• Natural conversation flow
• Extensive knowledge base

Specifications

• Context: 128K tokens
• Strengths: Versatility
• Speed: 30-40 tokens/sec
• Best for: General tasks
• Cost: Premium tier

GPT-5 access may be limited based on your API tier

GPT-4 Turbo

Production-Ready Workhorse

Optimized for real-world applications:

• Consistent performance
• Reliable outputs
• Good token efficiency
• Broad language support
• Stable API

Specifications

• Context: 128K tokens
• Strengths: Reliability
• Speed: 50-60 tokens/sec
• Best for: Production
• Cost: Standard tier

Google Models

Google Gemini Family

Massive context windows and multimodal capabilities

Gemini 2.5 Pro

The Context Champion

Unmatched context window for large-scale analysis:

• 1M+ token context window
• Excellent for large codebases
• Strong multimodal support
• Fast inference
• Competitive pricing

Specifications

• Context: 1M+ tokens
• Strengths: Large context
• Speed: 60-80 tokens/sec
• Best for: Big projects
• Cost: Competitive

Gemini 2.5 Flash

Lightning Fast Responses

Optimized for speed without sacrificing quality:

• Extremely fast inference
• Lower latency
• Cost-efficient
• Good for simple tasks
• Real-time applications

Specifications

• Context: 32K tokens
• Strengths: Speed
• Speed: 100+ tokens/sec
• Best for: Quick tasks
• Cost: Budget tier

Specialized Models

Purpose-Built for Specific Domains

Models optimized for particular use cases and languages

DeepSeek Coder

The Code Specialist

Trained specifically for programming tasks:

• Superior code completion
• Bug detection and fixing
• Code explanation
• Refactoring suggestions
• Multi-language support

Supported Languages: Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more

Performance Metrics

Use DeepSeek for intensive coding sessions and complex debugging

Qwen 2.5

Multilingual Excellence

Outstanding performance across languages:

• Native multilingual training
• Strong in Asian languages
• Excellent reasoning
• Cultural awareness
• Technical accuracy

Language Support: English, Chinese, Japanese, Korean, Spanish, French, German, and 50+ more

Use Cases

• International projects
• Localization tasks
• Cross-cultural apps
• Global documentation
• Multi-region deployment

Local Models

Privacy-First AI on Your Machine

Run powerful models locally without internet connection

Setup Guide

1. Download Model

   Model Sources

   Download llamafile models from:

   • Official Llamafile Repo
   • Hugging Face
   • Direct Links
2. Install Model

   # Create models directory
   mkdir -p ~/.siya/local-models
   
   # Move downloaded model
   mv ~/Downloads/model.llamafile ~/.siya/local-models/
   
   # Make executable
   chmod +x ~/.siya/local-models/model.llamafile
   
3. Configure Siya

   Model appears automatically in Siya’s model selector

4. First Run

   First run may take time to load model into memory

Local models require 8-64GB RAM depending on model size

Available Models

Performance

Local Model Performance

Performance depends on your hardware configuration

Hardware Requirements

Minimum Specs

• CPU: 8-core modern processor
• RAM: 16GB (8GB models)
• Storage: 50GB free
• OS: macOS 12+, Windows 10+

Recommended Specs

• CPU: Apple M2/M3 or equivalent
• RAM: 32-64GB
• GPU: Optional but helpful
• Storage: 100GB+ SSD

Optimization Tips

1. Enable GPU Acceleration

   export LLAMAFILE_GPU=1
   export CUDA_VISIBLE_DEVICES=0
   
2. Adjust Context Size

   export LLAMAFILE_CTX=4096  # Reduce for speed
   
3. Use Quantization

   Choose Q4 or Q5 quantized models for better performance

Model Selection Strategy

Choose the Right Model Every Time

Strategic guidance for optimal model selection

By Task Type

Software Architecture

Recommended: Claude Opus 4

Why Opus 4?

• Deep reasoning for design decisions
• Understands complex systems
• Excellent at trade-off analysis
• Can model entire architectures

Alternative: GPT-5 for broader perspective

Code Implementation

Recommended: DeepSeek Coder

Why DeepSeek?

• Specialized for code generation
• Superior completion accuracy
• Understands language idioms
• Fast and efficient

Alternative: Claude Sonnet 4 for general coding

Debugging & Analysis

Recommended: Claude Sonnet 4

Why Sonnet 4?

• Balanced analysis capabilities
• Good at finding subtle bugs
• Clear explanations
• Fast enough for iteration

Alternative: Local Qwen for privacy-sensitive code

Documentation

Recommended: GPT-4 Turbo

Why GPT-4 Turbo?

• Natural writing style
• Good at examples
• Consistent formatting
• Cost-effective

Alternative: Gemini Pro for large docs

By Project Size

By Constraints

Budget Conscious

1. Primary: Gemini Flash

   Fastest and most cost-effective for simple tasks

2. Complex Tasks: Claude Sonnet 4

   Good balance of capability and cost

3. Consider Local Models

   One-time download, no ongoing costs

Privacy First

Local Models Only

1. Qwen 2.5 Coder 32B - Best for coding
2. Llama 3.3 70B - General purpose
3. Mistral 7B - Resource-constrained

Never use cloud models for sensitive data

Speed Critical

Cloud Options

1. Gemini Flash
2. GPT-4 Turbo
3. Claude Sonnet 4

Local Options

1. Mistral 7B
2. Phi-3 Mini
3. Quantized models

Model Comparison

Side-by-Side Model Comparison

Detailed comparison to help you make informed decisions

Switching Models

Dynamic Model Switching

Switch models on the fly based on your needs

Manual Switching

1. Quick Switch

   Press ⌘ + M to open model selector

2. Menu Bar

   Click model name in status bar

3. Command

   Type “switch to GPT-5” in chat

4. Settings

   Change default in preferences

Automatic Switching

Smart Model Selection

Enable automatic model switching based on task

{
  "modelSelection": {
    "autoSwitch": true,
    "rules": [
      {
        "pattern": "debug|fix|error",
        "model": "claude-sonnet-4"
      },
      {
        "pattern": "architect|design|plan",
        "model": "claude-opus-4"
      },
      {
        "pattern": "quick|simple|fast",
        "model": "gemini-flash"
      },
      {
        "complexity": "> 8",
        "model": "claude-opus-4"
      }
    ]
  }
}

Siya learns your preferences and improves selection over time

Cost Optimization

Manage Your AI Budget Effectively

Strategies to optimize costs while maintaining quality

Cost-Saving Strategies

1. Use Appropriate Models

   Don’t use Opus 4 for simple tasks - Sonnet 4 or Flash work great

2. Enable Auto-Switching

   Let Siya choose cost-effective models automatically

3. Leverage Local Models

   Use local models for repetitive or sensitive tasks

4. Optimize Context

   Clear unnecessary context to reduce token usage

5. Batch Operations

   Group similar tasks to reduce overhead

Usage Monitoring

Monitor Commands

"Show my token usage"
"What's my API spend today?"
"Which model am I using most?"
"Show cost breakdown"

Best Practices

Model Selection Excellence

Expert tips for getting the most from each model

General Guidelines

Do's

• Start with faster models
• Upgrade when needed
• Use local for privacy
• Monitor costs
• Learn model strengths
• Experiment freely

Don'ts

• Always use most expensive
• Ignore context limits
• Share API keys
• Neglect local options
• Stick to one model
• Waste tokens

Model-Specific Tips

Future Models

Coming Soon to Siya

Upcoming models and capabilities on our roadmap

In Development

Coming Soon

• Claude Haiku 3
• GPT-4 Vision
• Llama 4
• Custom fine-tuned models
• Mixtral MOE

Future Vision

On the Horizon

• Multi-model ensemble
• Custom training
• Specialized verticals
• Real-time models
• Edge deployment

Summary

Choose Wisely, Code Brilliantly

With Siya’s comprehensive model selection, you have access to the perfect AI for every task. From the reasoning depth of Claude Opus 4 to the lightning speed of Gemini Flash, from specialized coding models to privacy-preserving local options - the right model is always at your fingertips. Understanding each model’s strengths empowers you to work smarter, faster, and more efficiently.

Get Started

Start using Siya with your chosen model

API Setup

Configure your API keys

---

The right model. The right time. The right results.