Exploring the Model Registry

RubyLLM ships with a built-in registry of 600+ models across all supported providers. The registry is bundled with the gem — no API calls or keys needed to browse it.

The Registry API

RubyLLM.models.all                    # every model
RubyLLM.models.chat_models            # chat/completion models
RubyLLM.models.embedding_models       # embedding models
RubyLLM.models.by_provider(:openai)   # filter by provider
RubyLLM.models.find("claude-sonnet-4-6")  # find specific model

Your Task

Open explore_models.rb and write code to:

1. Print the total number of models in the registry
2. List all providers (unique, sorted)
3. Print the first 10 chat models with their id, provider, and context window
4. Find a specific model and display its details

# 1. Total models
puts "Total models: #{RubyLLM.models.all.count}"

# 2. Providers
providers = RubyLLM.models.all.map(&:provider).uniq.sort
puts "Providers: #{providers.join(', ')}"

# 3. First 10 chat models
RubyLLM.models.chat_models.first(10).each do |model|
  puts "  %-30s %-12s %s tokens" % [model.id, model.provider, model.context_window]
end

# 4. Specific model details
model = RubyLLM.models.find("gpt-4o")
puts "  Context window: #{model.context_window} tokens"
puts "  Vision: #{model.supports_vision?}"

Run it:

$ ruby explore_models.rb

This works without any API keys — the data is bundled in the gem.

Model Properties

Each model exposes rich metadata:

Filtering with Enumerable

Since the registry returns arrays, you can chain Ruby’s Enumerable methods:

# Find all OpenAI models with vision support
RubyLLM.models.by_provider(:openai).select(&:supports_vision?)

# Find the cheapest chat model
RubyLLM.models.chat_models.min_by(&:input_price_per_million)

# Group models by provider
RubyLLM.models.all.group_by(&:provider).transform_values(&:count)

# Filter by model family
RubyLLM.models.by_family("claude3_sonnet")

Custom Endpoints & Unlisted Models

Some deployments use custom API endpoints (Azure OpenAI, local Ollama, etc.) or models not yet in the registry. Use assume_model_exists: true to bypass registry validation:

# Azure OpenAI endpoint
RubyLLM.configure do |config|
  config.openai_api_base = "https://YOUR_RESOURCE.openai.azure.com"
  config.openai_api_key  = ENV["AZURE_OPENAI_KEY"]
end

chat = RubyLLM.chat(
  model: "my-gpt4-deployment",
  provider: :openai,
  assume_model_exists: true
)

The provider: keyword is required when assuming existence — it tells RubyLLM which API format to use.

Keeping the Registry Fresh

The bundled registry is a snapshot. To pull the latest models from all configured providers:

RubyLLM.models.refresh!             # update in memory
RubyLLM.models.save_to_json         # persist to disk

In a Rails app, run this after initial setup. The rake models:update task is for gem maintainers only — use refresh! in application code.