Skip to main content

LLM API Reference

Overview​

The CAIP LLM API provides unified access to multiple large language model (LLM) providers and embedding models. It enables you to interact with commercial and self-hosted LLMs for chat, embedding, image generation, video generation, audio, and responses use cases via a consistent REST interface.

  • Base URL (RoW): https://llm.api.caip.bmw.cloud
  • Base URL (China): https://llm.api.caip.bmwchina.cloud

Authentication​

All endpoints require an API key via the Authorization header.

RoW (new stack) — requires the standard OAuth2 Bearer token format:

Authorization: Bearer <your-api-key>

China (new stack) — uses the same OAuth2 Bearer token format as RoW:

Authorization: Bearer <your-api-key>
warning

Only UUID-format API keys are accepted by the RoW endpoint. Old-format keys are deprecated. Request new keys via the Self-Service Portal.

Endpoints​

Chat Completions​

POST /v1/chat/completions

Initiate a chat with your preferred LLM model backend.

Request Body:

The model must be specified in the JSON body via the model parameter.

{
  "model": "gpt-4o",
  "messages": [
    { "role": "system", "content": "You are a mathematician" },
    { "role": "user", "content": "What is 1+1?" }
  ]
}

Response:

{
  "choices": [
    {
      "index": 0,
      "finish_reason": "stop",
      "message": {
        "role": "assistant",
        "content": "As a mathematician, I can confidently say that 1+1 = 2."
      }
    }
  ],
  "object": "chat.completion",
  "model": "gpt-4o-2024-08-06",
  "usage": {
    "completion_tokens": 42,
    "prompt_tokens": 12,
    "total_tokens": 54
  }
}

Models Supported (RoW):

  • AWS Bedrock: claude-haiku-4.5, claude-sonnet-4.6, claude-opus-4.5, claude-opus-4.6, nova-lite, nova-micro, nova-pro, deepseek-r1, deepseek-v3, deepseek-v3.1, qwen-plus, qwen-flash, qwen3-30b-a3b-instruct-2507
  • Azure OpenAI: gpt-4o, gpt-4o-mini, gpt-41, gpt-41-mini, gpt-41-nano, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-2, gpt-5.5, o3

Realtime​

GET wss://llm.api.caip.bmwchina.cloud/v1/realtime?model={model}

OpenAI Realtime-compatible WebSocket endpoint for bidirectional text and audio conversations in China. Authenticate the handshake with Authorization: Bearer {CAIP_API_KEY}.

Models Supported (China): qwen3.5-omni-plus-realtime, qwen3.5-omni-flash-realtime

Send OpenAI Realtime events such as session.update, conversation.item.create, and response.create. See the full realtime guide for a working example and audio requirements.

Responses​

POST /v1/responses

OpenAI's advanced interface for generating model responses with stateful interactions. Supports text and image inputs.

Models Supported (RoW): gpt-5-codex, gpt-5-pro, o3-pro

Embeddings​

POST /v1/embeddings

Generate vector embeddings from text input.

Request Body:

{
  "model": "text-embedding-3-small",
  "input": "Introduction to embedding models",
  "dimensions": 1024,
  "encoding_format": "float"
}

Response:

{
  "data": [
    {
      "embedding": [ -0.033, -0.015, ... ],
      "index": 0,
      "object": "embedding"
    }
  ],
  "object": "list",
  "model": "text-embedding-3-small",
  "usage": {
    "prompt_tokens": 6,
    "total_tokens": 6
  }
}

Models Supported (RoW):

  • Azure: text-embedding-3-small, text-embedding-3-large
  • AWS Bedrock: titan-text-embeddings-v2

Image Generation​

POST /v1/images/generations

Generate images from text prompts.

Models Supported (RoW): gpt-image-1, gpt-image-1-mini, gpt-image-1.5, gpt-image-2, titan-image-generator-v2

Video Generation​

Video generation is asynchronous — submit a job, poll for completion, then download the result.

StepMethodEndpointDescription
1POST/v1/videosCreate a video generation job
2GET/v1/videos/{video_id}Poll until status is "completed"
3GET/v1/videos/{video_id}/contentDownload the generated video file

Models Supported (RoW): sora-2, wan-video

Audio Transcription​

POST /v1/audio/transcriptions

Transcribe audio files to text.

Models Supported (RoW): gpt-4o-transcribe, gpt-4o-mini-transcribe, whisper (deprecated)

Audio Speech (TTS)​

POST /v1/audio/speech

Generate speech from text.

Models Supported (RoW): gpt-4o-mini-tts

Error Codes​

  • 400: Invalid input (e.g., missing model parameter, malformed request)
  • 401: Unauthorized (invalid API key, missing Bearer prefix, expired key)
  • 422: Validation exception
  • 429: Rate limit exceeded

External Documentation​

Contact​

For support, contact: ConnectedAI@bmw.de

License: Apache 2.0