[FEATURE] configure model deployments with non standard names to use the standard name for cost and capabilities discovery

Scope check

• This is core LLM communication (not application logic)
• This benefits most users (not just my use case)
• This can't be solved in application code with current RubyLLM
• I read the Contributing Guide

Due diligence

• I searched existing issues
• I checked the documentation

What problem does this solve?

I have an azure model deployment named gpt-4o-global. This is a deployment of gpt-4o. It would be a nice feature to be able to create something like an alias (with a similar json file) that maps gpt-4o-global to gpt-4o, so that cost and capabilities of the model are known when a new Chat is created in the database.

Proposed solution

file lib/ruby_llm/cost_and_capability_mapping.json

{
  'got-4o-global': 'gpt-4o',
}

create a new Chat with gpt-4o-global with provider and assume_model_exists: true still uses the mapping for the model capabilities

chat = Chat.create(model: 'gpt-4o-global', provider: :azure, assume_model_exists: true)
chat.model
#<RubyLLM::ActiveRecord::Model:0x0000ffff6ab75650
 id: 1877,
 capabilities: ["function_calling", "structured_output", "vision", "tool_choice", "parallel_tool_calls"],
 context_window: 128000,
 created_at: "2026-09-17 17:36:31.195829000 -0400",
 family: "gpt",
 knowledge_cutoff: nil,
 max_output_tokens: "[FILTERED]",
 metadata:
  {"cost" => {"input" => 2.5, "output" => 10, "cache_read" => 1.25}, "limit" => {"output" => 16384, "context" => 128000}, "object" => "model", "source" => "models.dev", "owned_by" => "system", "knowledge" => "2023-09", "reasoning" => false, "tool_call" => true, "attachment" => true, "provider_id" => "openai", "temperature" => true, "last_updated" => "2024-08-06", "open_weights" => false, "structured_output" => true},
 modalities: {"input" => ["text", "image", "pdf"], "output" => ["text"]},
 model_created_at: "2024-05-12 20:00:00.000000000 -0400",
 model_id: "gpt-4o-global",
 name: "GPT 4o global",
 pricing: {"text_tokens" => {"standard" => {"input_per_million" => 2.5, "output_per_million" => 10, "cache_read_input_per_million" => 1.25}}},
 provider: "openai",
 unlisted_at: nil,
 updated_at: "2026-09-17 17:36:31.195829000 -0400">

instead of

#<RubyLLM::ActiveRecord::Model:0x0000ffff8256ba58
 id: 1817,
 capabilities: ["function_calling", "streaming", "vision", "structured_output"],
 context_window: nil,
 created_at: "2026-09-17 17:38:33.123566000 -0400",
 family: nil,
 knowledge_cutoff: nil,
 max_output_tokens: nil,
 metadata: {"warning" => "Assuming model exists, capabilities may not be accurate"},
 modalities: {"input" => ["text", "image"], "output" => ["text"]},
 model_created_at: nil,
 model_id: "gpt-4o-global",
 name: "Gpt 4o global",
 pricing: {},
 provider: "azure",
 unlisted_at: "2026-09-17 17:38:54.347494000 -0400",
 updated_at: "2026-09-17 17:38:54.349078000 -0400">

Why this belongs in RubyLLM

This would make costs stored in the database in messages more accurate. It would also help the harness know what actual capabilities a model has when using it.

Start with lib/ruby_llm/cost_and_capability_mapping.json and trace model creation when assume_model_exists is true. Implement the proposed alias mapping so a deployment such as gpt-4o-global uses the standard model's cost and capabilities while retaining its deployment model_id. Done means the stored model data reflects the mapped model rather than the fallback warning and empty pricing.