List models

Returns the Token Factory model list, with the pricing and capability metadata for each entry. A model service on your instance serves only the model you launched it with, and answers this path from vLLM instead. See Model service endpoints.

GET /v1/models API key or session

Also reachable at /api/v1/models.

Request

No parameters.

curl https://developer.amd.com.cn/radeon/api/v1/models \
  -H "Authorization: Bearer $RADEON_API_KEY"

Response

The envelope is a bare { "data": [...] }. There is no object field, and entries carry richer metadata than OpenAI’s model object, without owned_by or created.

{
  "data": [
    {
      "id": "DeepSeek-V4-Flash",
      "name": "DeepSeek-V4-Flash",
      "aliases": [],
      "description": "Radeon DeepSeek V4 Flash served by the AMD GPU Cloud",
      "family": "deepseek",
      "architecture": {
        "input_modalities": ["text"],
        "output_modalities": ["text"],
        "tokenizer": "GPT"
      },
      "top_provider": { "is_moderated": true },
      "providers": [
        {
          "providerId": "radeon-deepseek",
          "externalId": "DeepSeek-V4-Flash",
          "pricing": {
            "prompt": "0.00000014",
            "completion": "0.00000028",
            "input_cache_read": "0.0000000028"
          },
          "streaming": true,
          "vision": false,
          "tools": true,
          "reasoning": true,
          "stability": "stable"
        }
      ],
      "pricing": {
        "prompt": "0.00000014",
        "completion": "0.00000028",
        "input_cache_read": "0.0000000028"
      },
      "context_length": 1048576,
      "supported_parameters": [
        "temperature", "max_tokens", "top_p",
        "frequency_penalty", "presence_penalty",
        "stream", "response_format", "tools"
      ],
      "json_output": true,
      "structured_outputs": true,
      "free": true,
      "stability": "stable"
    }
  ]
}

Use any id from this list as the model field in chat completions.

Fields

Field	Description
id	The name to send as model.
name, description, family, aliases	Display metadata.
architecture	Input and output modalities, and the tokenizer family.
providers	The backends serving this model, each with its own pricing and capability flags.
pricing	Price per token as decimal strings: prompt, completion, and input_cache_read where prompt caching applies. A model billed by input characters, such as OmniVoice, puts its price per character in prompt. Counted against your daily allowance in Token Factory points, not charged as money. See Usage and quota.
context_length	Maximum context window in tokens.
supported_parameters	Request parameters this model accepts. The full schema is under chat completions.
json_output, structured_outputs	Whether response_format is honoured.
free	Whether calls are billed against your daily allowance at zero cost.
stability	stable, beta, unstable, or experimental.

Not OpenAI’s model object

Clients that expect data[].object == "model" or read owned_by will not find those fields. Everything else in Token Factory follows OpenAI’s schema; this endpoint does not.

Notes

The catalog changes. Models are added and retired over time, so resolve the list at runtime rather than hard-coding names, and handle the case where a model you used yesterday has gone.

The catalog is the same for every key. It isn’t scoped to your account.

This endpoint doesn’t count against your concurrency allowance, though the per-minute rate limit still applies. It’s cheap to call, but cache the result for a few minutes rather than calling it before every completion.

The same metadata is rendered on each model’s card in the Token Factory.