Models and prices

Ballast serves a small catalog of open models. Each host runs them through Ollama on its own GPU, and you pick one with the model field. A model name that is not in the catalog returns 404 model_not_found.

Prices are per million tokens, input then output. You pay for the tokens that were delivered, at these prices. Every model has a 4,096 token context. For how many hosts have each model loaded right now and how fast the first token arrives, see Models and prices.

List models from the API

curl https://ballast.outerscope.xyz/v1/models -H "Authorization: Bearer $BALLAST_API_KEY"

{
  "object": "list",
  "data": [
    { "id": "llama-3.1-8b-instruct", "object": "model", "created": 1794096000, "owned_by": "ballast", "context_length": 4096 }
  ]
}

The list holds every catalog model, one entry each. The call needs a valid key, like the rest of the API.

Which one to pick

• llama-3.1-8b-instruct is the cheapest and the quickest. Start here for chat, classification and anything you will call a lot.
• qwen3-14b costs more and is a step up in answers. A good default when quality matters.
• qwen3-32b is the biggest. Fewer hosts have the VRAM for it, so it can be scarcer at busy times.

Ballast is not the cheapest place for these models. The prices are set so a host's electricity is covered. If you want steady capacity for one model, rent a whole GPU by the hour.