Available Models

Mixpeek offers three tiers of embedding model implementations, ranging from simplified multimodal options to custom enterprise solutions.

​

1. Simplified Multimodal Models

These models provide an easy-to-use interface for common modalities. Mixpeek automatically selects and updates the most performant models based on industry benchmarks.

Example usage:

{
    "type": "text",
    "value": "search query",
    "embedding_model": "text"  # Uses the default model
}

​

2. Hosted Model Selection

For more control, you can specify exact model implementations:

Example usage:

{
    "type": "text",
    "value": "search query",
    "embedding_model": "baai-bge-m3"  # Specific model selection
}

​

3. Custom Model Integration

Enterprise customers can integrate their own models:

1. Deploy your model with a compatible API
2. Register the model in your namespace
3. Use your custom model ID in requests

Example configuration:

{
    "namespace": "enterprise-ns",
    "model_config": {
        "id": "custom-model-v1",
        "endpoint": "https://your-model-endpoint",
        "supported_modalities": ["text", "image"],
        "vector_size": 768
    }
}

Example usage:

# X-Namespace Header: enterprise-ns
{
    "type": "text",
    "value": "search query",
    "embedding_model": "custom-model-v1",
}

​

Model Registry Details

Each model in the registry includes:

• Supported modalities
• Vector type (dense or sparse)
• Output dimensions
• Hosting location and provider
• Performance metrics