Available Models¶

AIA supports a wide range of AI models through the RubyLLM gem. This comprehensive list shows all supported models, their capabilities, and best use cases.

Viewing Available Models¶

Command Line Query¶

# List all available models
aia --available_models

# Filter by provider
aia --available_models openai
aia --available_models anthropic
aia --available_models google

# Filter by capability
aia --available_models vision
aia --available_models function_calling
aia --available_models text_to_image

# Complex filtering (AND operation)
aia --available_models openai,gpt,4
aia --available_models anthropic,claude,sonnet

Within Prompts¶

# List models in a prompt
//available_models

# Filter models
//available_models openai,gpt

Model Categories¶

OpenAI Models¶

GPT-4 Family¶

gpt-4: Most capable model, excellent for complex reasoning
Context: 8,192 tokens
Best for: Complex analysis, creative writing, code generation
Cost: Higher, but highest quality
gpt-4-turbo: Faster GPT-4 with larger context
Context: 128,000 tokens
Best for: Long documents, comprehensive analysis
Cost: Lower than GPT-4, faster responses
gpt-4-vision-preview: GPT-4 with image understanding
Context: 128,000 tokens (including images)
Best for: Image analysis, visual content creation
Capabilities: Text + image input, text output

GPT-3.5 Family¶

gpt-3.5-turbo: Fast, cost-effective general purpose
Context: 4,096 tokens
Best for: General queries, quick tasks, batch processing
Cost: Most economical
gpt-3.5-turbo-16k: Extended context version
Context: 16,384 tokens
Best for: Longer documents, extended conversations
Cost: Moderate

Specialized OpenAI Models¶

text-davinci-003: Legacy completion model
code-davinci-002: Code-optimized model
text-embedding-ada-002: Text embedding model

Anthropic Claude Models¶

Claude-3 Family¶

claude-3-opus: Highest capability Claude model
Context: 200,000 tokens
Best for: Complex analysis, long documents, nuanced tasks
Cost: Premium pricing
claude-3-sonnet: Balanced performance and cost
Context: 200,000 tokens
Best for: Most general tasks, good balance
Cost: Moderate
claude-3-haiku: Fastest, most economical
Context: 200,000 tokens
Best for: Quick tasks, batch processing, simple queries
Cost: Most economical

Claude-2 Family (Legacy)¶

claude-2: Previous generation
Context: 100,000 tokens
Best for: Long-form content, analysis
Status: Being phased out

Google Models¶

Gemini Family¶

gemini-pro: Google's flagship model
Context: 32,000 tokens
Best for: Reasoning, structured data, math
Features: Multimodal capabilities
gemini-pro-vision: Gemini with vision
Context: 32,000 tokens (including images)
Best for: Image understanding, visual analysis
Capabilities: Text + image input

PaLM Family¶

text-bison: Text generation model
chat-bison: Conversational model

Open Source Models (via Ollama)¶

Llama 2 Family¶

llama2-7b: 7 billion parameter model
Best for: Local deployment, privacy-sensitive tasks
Requirements: 8GB+ RAM
llama2-13b: 13 billion parameter model
Best for: Better quality local processing
Requirements: 16GB+ RAM
llama2-70b: 70 billion parameter model
Best for: Highest quality local processing
Requirements: 64GB+ RAM

Code Llama¶

codellama-7b: Code-specialized 7B model
codellama-13b: Code-specialized 13B model
codellama-34b: Code-specialized 34B model

Other Open Models¶

mistral-7b: Efficient general-purpose model
mixtral-8x7b: Mixture of experts model
phi-2: Microsoft's compact model
orca-2: Microsoft's reasoning-focused model

Model Capabilities¶

Text Generation¶

All models support: Basic text generation, question answering, summarization

Best performers: - Complex reasoning: GPT-4, Claude-3-Opus - Creative writing: GPT-4, Claude-3-Sonnet - Technical writing: Claude-3-Sonnet, GPT-4

Code Understanding and Generation¶

Code-optimized models: - CodeLlama family (7B, 13B, 34B) - GPT-4 (excellent general code understanding) - Claude-3-Sonnet (good at following coding standards)

Capabilities: - Code generation and completion - Bug detection and fixing - Code explanation and documentation - Refactoring suggestions

Vision and Multimodal¶

Image understanding models: - GPT-4 Vision Preview - Gemini Pro Vision - Claude-3 (limited vision capabilities)

Capabilities: - Image description and analysis - Chart and diagram interpretation - OCR and text extraction - Visual question answering

Function Calling and Tools¶

Tool-compatible models: - GPT-3.5-turbo (excellent function calling) - GPT-4 (sophisticated tool usage) - Claude-3-Sonnet (good tool integration)

Use cases: - API integrations - Database queries - File system operations - External service calls

Choosing the Right Model¶

By Task Type¶

Quick Tasks and Batch Processing¶

# Fast, economical models
aia --model gpt-3.5-turbo simple_task
aia --model claude-3-haiku batch_processing

Complex Analysis and Reasoning¶

# High-capability models
aia --model gpt-4 complex_analysis
aia --model claude-3-opus comprehensive_research

# Code-optimized models
aia --model codellama-34b code_generation
aia --model gpt-4 code_review

Long Documents¶

# Large context models
aia --model claude-3-sonnet long_document.pdf
aia --model gpt-4-turbo comprehensive_analysis.md

Image Analysis¶

# Vision-capable models
aia --model gpt-4-vision-preview image_analysis.jpg
aia --model gemini-pro-vision chart_interpretation.png

By Budget Considerations¶

Cost-Effective Options¶

gpt-3.5-turbo: Best general-purpose budget option
claude-3-haiku: Anthropic's economical choice
Local models: Ollama-based models (compute cost only)

Premium Options¶

gpt-4: OpenAI's flagship
claude-3-opus: Anthropic's highest capability
gpt-4-turbo: Large context with good performance

By Privacy and Security¶

Cloud-Based (Standard)¶

OpenAI models (GPT-3.5, GPT-4)
Anthropic models (Claude-3 family)
Google models (Gemini family)

Local/Self-Hosted¶

Ollama models (Llama 2, CodeLlama, Mistral)
Privacy-focused deployment
Full control over data

Model Configuration Examples¶

Development Workflow¶

# Different models for different stages
development:
  quick_tasks: gpt-3.5-turbo
  code_review: gpt-4
  documentation: claude-3-sonnet
  testing: codellama-13b

Content Creation Workflow¶

content:
  research: claude-3-sonnet
  drafting: gpt-4
  editing: claude-3-opus
  seo_optimization: gpt-3.5-turbo

Analysis Workflow¶

analysis:
  data_exploration: claude-3-sonnet
  statistical_analysis: gemini-pro
  insights: gpt-4
  reporting: claude-3-haiku

Model Performance Comparison¶

Speed (Responses per minute)¶

gpt-3.5-turbo: ~60 RPM
claude-3-haiku: ~50 RPM
gemini-pro: ~40 RPM
gpt-4: ~20 RPM
claude-3-opus: ~15 RPM

Context Window Size¶

Claude-3 family: 200,000 tokens
GPT-4-turbo: 128,000 tokens
Gemini-pro: 32,000 tokens
GPT-3.5-turbo-16k: 16,384 tokens
GPT-4: 8,192 tokens

Cost Efficiency (approximate)¶

gpt-3.5-turbo: Most economical
claude-3-haiku: Very economical
gemini-pro: Moderate
claude-3-sonnet: Moderate-high
gpt-4: Premium
claude-3-opus: Most expensive

Advanced Model Usage¶

Multi-Model Strategies¶

# Use different models for different aspects
aia --model gpt-3.5-turbo initial_analysis.txt
aia --model gpt-4 --include initial_analysis.txt detailed_review.txt
aia --model claude-3-sonnet --include detailed_review.txt final_synthesis.txt

Model Switching Based on Content¶

# Dynamic model selection
//ruby
content_size = File.read('<%= input %>').length
complexity = content_size > 10000 ? 'high' : 'low'

model = case complexity
        when 'high' then 'claude-3-sonnet'
        when 'low' then 'gpt-3.5-turbo'
        end

puts "//config model #{model}"

Fallback Strategies¶

# Model fallback chain
//ruby
preferred_models = ['gpt-4', 'claude-3-sonnet', 'gpt-3.5-turbo']
available_models = `aia --available_models`.split("\n").map { |line| line.split.first }

selected_model = preferred_models.find { |model| available_models.include?(model) }
puts "//config model #{selected_model || 'gpt-3.5-turbo'}"

Staying Current¶

Model Updates¶

Check regularly: aia --available_models
Version changes: Models are updated periodically
New releases: Follow provider announcements
Deprecations: Some models may be retired

Performance Monitoring¶

# Test model performance
time aia --model gpt-4 test_prompt
time aia --model claude-3-sonnet test_prompt

# Compare outputs
aia --model "gpt-4,claude-3-sonnet" --no-consensus comparison_test

Working with Models - Model selection and configuration
Configuration - Model configuration options
CLI Reference - Model-related command-line options
Chat Mode - Interactive model usage
Advanced Prompting - Model-specific techniques

The AI landscape evolves rapidly. Regularly check for new models and updates to ensure you're using the best tools for your specific needs!