Skip to main content
GET
List language models
Returns the current Magica language-model catalog. Use it to populate a model picker or validate a configured model ID before sending an inference request. This endpoint does not require authentication; inference requests do. The catalog contains billable models that can produce synchronous text or tool results through the LLM Gateway. It does not list asynchronous batch-only variants or model types that require different APIs, such as embeddings, reranking, speech, transcription, image generation, or video generation. Every listed model ID is selectable on Chat Completions, Messages, and Responses. Choose the endpoint for its wire format, then use each model’s capabilities and limits to decide whether it fits your input.

Request

Use the exact key or id from magica.models as the model value in an inference request. The catalog changes as models become available or are retired, so clients should not hard-code the complete list.

What each model includes

  • Public model ID used in model
  • Display name and current availability
  • Input capabilities such as text and images
  • Context and maximum output-token limits
  • limits.request_bytes, the maximum final encoded request size
  • Input, output, cache-read, and cache-write prices per one million tokens
Capabilities and limits vary by model. Check the live catalog instead of assuming every model accepts images, PDFs, tools, or the same context size.

Request limits

limits.request_bytes is measured after the request has been encoded for the selected model. Requests above that value return 413; Magica does not silently remove conversation history or attachments.

Pricing units

cost.input, cost.output, cost.cache_read, and cost.cache_write are US-dollar prices per one million tokens. Successful inference responses report the token usage used for credit settlement.