ragflow/conf/models/deepinfra.json

{
  "name": "DeepInfra",
  "url": {
    "default": "https://api.deepinfra.com"
  },
  "url_suffix": {
    "chat": "v1/chat/completions",
    "models": "models/list",
    "balance": "payment/checklist",
    "rerank": "v1/inference",
    "embedding": "v1/embeddings",
    "tts": "v1/text-to-speech",
    "asr": "v1/audio/transcriptions"
  },
  "class": "deepinfra",
  "models": [
    {
      "name": "deepseek-ai/DeepSeek-V3.2",
      "max_tokens": 32768,
      "model_types": [
        "chat"
      ],
      "thinking": {
        "default_value": true,
        "clear_thinking": true
      }
    },
    {
      "name": "Qwen/Qwen3-Embedding-4B",
      "max_tokens": 8192,
      "model_types": [
        "embedding"
      ]
    },
    {
      "name": "hexgrad/Kokoro-82M",
      "max_tokens": 16384,
      "model_types": [
        "tts"
      ]
    },
    {
      "name": "bosonai/HiggsAudioV2.5",
      "max_tokens": 8192,
      "model_types": [
        "asr"
      ]
    }
  ]
}