DeepInfra

DeepInfra provides access to open-source embedding models with OpenAI-compatible API.

Overview

Models: 16 models (Qwen3, BGE, E5, GTE, and more)
Key Features: Open-source models, Qwen3 Matryoshka support, competitive pricing
API Docs: DeepInfra Embeddings

Environment Variable

export DEEPINFRA_API_KEY="your-deepinfra-api-key"

Supported Parameters

Parameter	Type	Required	Description
`model`	str	Yes	Model identifier
`input`	str \| List[str]	Yes	Text(s) to embed
`dimensions`	int	No	Output dimensions (Qwen3 models only, 32-8192)
`api_key`	str	No	Override API key

Note: DeepInfra does not support input_type parameter.

Examples

Basic Usage

response = client.embed(
    model="Qwen/Qwen3-Embedding-8B",
    input="Hello, DeepInfra!"
)

With Dimensions (Qwen3 models)

# Qwen3 models support Matryoshka (32-8192)
response = client.embed(
    model="Qwen/Qwen3-Embedding-8B",
    input="Sample text",
    dimensions=1024  # vs default 4096
)

print(f"Dimensions: {response.dimensions}")  # 1024

Different Model Sizes

# Large model (8B parameters, 4096 dimensions)
large_response = client.embed(
    model="Qwen/Qwen3-Embedding-8B",
    input="Text"
)

# Medium model (4B parameters, 2560 dimensions)
medium_response = client.embed(
    model="Qwen/Qwen3-Embedding-4B",
    input="Text"
)

# Small model (0.6B parameters, 1024 dimensions)
small_response = client.embed(
    model="Qwen/Qwen3-Embedding-0.6B",
    input="Text"
)

Special Notes

⚠️ input_type is NOT supported
✅ Matryoshka dimensions (32-8192) for Qwen3 models only
OpenAI-compatible endpoint
Open-source models (BAAI, Qwen, UAE, etc.)
Model names use provider prefix (e.g., Qwen/, BAAI/)

Next Steps

Models Catalog - View all DeepInfra models
Common Parameters: dimensions - Matryoshka embeddings

On this page