BGE-M3 Embedding TL5

TL5 §203Open weights

Multilingual embeddings for hybrid search and RAG across 100+ languages and 8K tokens

Pricing

Prices for this trust level. Input images are priced per MPx and input text per 1M tokens. Output is not priced.

Trust levels for this model
Trust Level Region Context Max outputInput image / MPxInput token / 1M§203
TL5 DE 8K — - - 0.01 € 0.01 $ §203

USD prices are converted from EUR at the ECB reference rate of 1 EUR = 1.1014 USD (2026-11-16). EUR prices are authoritative.

Prices as of 16.11.2026.

About this model

BGE-M3 (BAAI General Embedding M3) is an open-weight text embedding model developed by the Beijing Academy of Artificial Intelligence. Built on XLM-RoBERTa with 568M parameters, it supports more than 100 languages and inputs up to 8,192 tokens, from short queries to long documents. Its dense embeddings have 1,024 dimensions. Unlike embeddings limited to a single retrieval method, BGE-M3 supports dense semantic search, sparse lexical matching, and ColBERT-style multi-vector retrieval within one model. These complementary representations enable hybrid search that combines meaning with exact-term matching. Typical applications include multilingual and cross-lingual search, enterprise knowledge-base retrieval, document discovery, and retrieval-augmented generation (RAG). The model is released under the MIT license.

API

Model ID bge-m3 OpenAI compatible

Base URL https://api.aki.io/v1

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_AKI_API_KEY",
    base_url="https://api.aki.io/v1",
)

response = client.embeddings.create(
    model="bge-m3",
    input="Hello!",
)
print(response.data[0].embedding)
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_AKI_API_KEY",
  baseURL: "https://api.aki.io/v1",
});

const response = await client.embeddings.create({
  model: "bge-m3",
  input: "Hello!",
});
console.log(response.data[0].embedding);
curl https://api.aki.io/v1/embeddings \
  -H "Authorization: Bearer $AKI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bge-m3",
    "input": "Hello!"
  }'

Frequently Asked Questions

How much does BGE-M3 Embedding TL5 cost?
Input: 0.01 € per 1M tokens. Prices are exclusive of VAT. Output is not priced.
What is the context window of BGE-M3 Embedding TL5?
BGE-M3 Embedding TL5 supports up to 8K tokens.
Which trust levels are available for BGE-M3 Embedding TL5?
BGE-M3 Embedding TL5 is available at TL5 (Germany Sovereign, DE).
Is BGE-M3 Embedding TL5 open weights?
Yes. The model weights are available on Hugging Face under the linked license.
Is BGE-M3 Embedding TL5 available with zero data retention?
Yes. BGE-M3 Embedding TL5 is served at TL5 (Germany Sovereign) with zero data retention and no model training.
How do I use BGE-M3 Embedding TL5 with the OpenAI SDK?
Use the model ID bge-m3 with the OpenAI-compatible endpoint at https://api.aki.io/v1.

Start Building with BGE-M3 Embedding TL5