Working with Models

RubyLLM provides a clean interface for discovering and working with AI models from multiple providers. This guide explains how to find, filter, and select the right model for your needs.

Exploring Available Models

RubyLLM automatically discovers available models from all configured providers:

# Get all available models
all_models = RubyLLM.models.all

# See how many models are available
puts "Total models: #{all_models.count}"

# List models with details
all_models.each do |model|
  puts "#{model.id} (#{model.provider}) - #{model.display_name}"
end

Filtering Models

You can filter models by various criteria:

# Get only chat models
chat_models = RubyLLM.models.chat_models

# Get only embedding models
embedding_models = RubyLLM.models.embedding_models

# Get only image generation models
image_models = RubyLLM.models.image_models

# Get only audio models
audio_models = RubyLLM.models.audio_models

Finding Models by Provider

Filter models by provider:

# Get OpenAI models
openai_models = RubyLLM.models.by_provider('openai')

# Get Anthropic models
anthropic_models = RubyLLM.models.by_provider('anthropic')

# Get Google models
google_models = RubyLLM.models.by_provider('gemini')

# Get DeepSeek models
deepseek_models = RubyLLM.models.by_provider('deepseek')

Chaining Filters

You can chain multiple filters to find exactly what you need:

# Get OpenAI chat models
openai_chat_models = RubyLLM.models.by_provider('openai').chat_models

# Get Anthropic embedding models (if any)
anthropic_embeddings = RubyLLM.models.by_provider('anthropic').embedding_models

# Get Google models that support vision
google_vision_models = RubyLLM.models.by_provider('gemini').chat_models.select(&:supports_vision)

Filters can be applied in any order:

# These return the same models
chat_anthropic = RubyLLM.models.chat_models.by_provider('anthropic')
anthropic_chat = RubyLLM.models.by_provider('anthropic').chat_models

Getting Model Details

Find a specific model by ID to see its capabilities:

model = RubyLLM.models.find('gpt-4o-mini')

puts "Model: #{model.display_name}"
puts "Provider: #{model.provider}"
puts "Context window: #{model.context_window} tokens"
puts "Max generation: #{model.max_tokens} tokens"
puts "Input price: $#{model.input_price_per_million} per million tokens"
puts "Output price: $#{model.output_price_per_million} per million tokens"
puts "Supports vision: #{model.supports_vision}"
puts "Supports functions: #{model.supports_functions}"
puts "Supports JSON mode: #{model.supports_json_mode}"

Using Enumerable Methods

The models collection includes Ruby’s Enumerable module, so you can use all your favorite methods:

# Count models by provider
provider_counts = RubyLLM.models.group_by(&:provider).transform_values(&:count)
puts "OpenAI models: #{provider_counts['openai']}"
puts "Anthropic models: #{provider_counts['anthropic']}"

# Find models matching specific criteria
vision_models = RubyLLM.models.select(&:supports_vision)
function_models = RubyLLM.models.select(&:supports_functions)

# Find the cheapest model for a task
cheapest_chat = RubyLLM.models.chat_models.min_by(&:input_price_per_million)
puts "Cheapest chat model: #{cheapest_chat.display_name} ($#{cheapest_chat.input_price_per_million}/M tokens)"

Finding Models by Family

Filter models by family (model architecture):

# Get all GPT-4 models
gpt4_models = RubyLLM.models.by_family('gpt4')

# Get all Claude 3 models
claude3_models = RubyLLM.models.by_family('claude3_sonnet')

Refreshing Model Data

Force a refresh of model data from providers:

RubyLLM.models.refresh!

Best Practices

When selecting models for your application:

  1. Consider context windows - Larger context windows support longer conversations but may cost more
  2. Balance cost vs. quality - More capable models cost more but may give better results
  3. Check capabilities - Make sure the model supports features you need (vision, functions, etc.)
  4. Use appropriate model types - Use embedding models for vector operations, chat models for conversations