Developers / Docs / Models

Models

Every request names a model by its exact ARK id. GET /api/v1/models lists the models that are online right now, with what each can do.

API reference: Models and status

List the models

The listing needs no API key.

curl https://api.ark-labs.cloud/api/v1/models

for model in client.models.list():
    print(model.id, model.ark_type, model.max_tokens)

for await (const model of client.models.list()) {
  console.log(model.id, model.ark_type, model.max_tokens);
}

Response (excerpt)

{
  "object": "list",
  "data": [
    {
      "id": "arklabsai/Qwen3.6-35B-A3B",
      "object": "model",
      "created": 1754035200,
      "owned_by": "ark",
      "shutdown_date": null,
      "ark_type": "multimodal",
      "max_tokens": 262144,
      "ark_reasoning": true
    },
    {
      "id": "arklabsai/bge-m3",
      "object": "model",
      "created": 1754035200,
      "owned_by": "ark",
      "shutdown_date": null,
      "ark_type": "embedding",
      "max_tokens": 32768,
      "ark_reasoning": false
    }
  ]
}

ARK fields

ark_typestringARK extension

What the model does, and so which endpoint serves it: causal_lm (text chat), multimodal (chat with image input), embedding or speech_transcription. Filter with ?type=multimodal.

max_tokensintegerARK extension

The model's context length on this deployment: the most tokens one request or stateful session can hold, prompt and output together.

ark_reasoningbooleanARK extension

Whether the model can think before answering (see Reasoning). Missing for a short while after the service starts.

A model appears in the list once it is fully online. If a model you used a moment ago is missing, it is being restarted: retry shortly.

Which model for which job

Model	Endpoint	Good for
`arklabsai/Qwen3.5-122B-A10B`	chat	The hardest tasks: multi-step reasoning, long documents, complex tool use, images.
`arklabsai/Qwen3.6-35B-A3B`	chat	The default choice: fast, capable, reasoning, tools and vision at a low price.
`arklabsai/Qwen3.5-4B`	chat	High-volume simple work: classification, extraction, routing, image tagging.
`arklabsai/Bielik-11B-v3.0-Instruct`	chat	Polish-language text, with a 32K context. No tool calls.
`arklabsai/bge-m3`	embeddings	Multilingual search and retrieval.
`arklabsai/whisper-large-v3-turbo`	transcriptions	Speech to text in 99 languages.

The model catalogue has the details of each model, its licence, and the models available for self-hosted deployments. Prices are on the pricing page.

Model ids

Use the full id including the arklabsai/ prefix. ARK has no aliases: gpt-4o or Qwen/Qwen3.6-35B-A3B return 404 model_not_found.
Calling a model on the wrong endpoint, such as chat with an embedding model, also returns 404.
GET /api/v1/models/{id} returns one model; the id contains a slash and is used as is.