We are now running two of the strongest multilingual open source retrieval models 2025/26 fully managed via our API. This adds two important components to the NWS Managed AI Models.
Anyone who has ever worked with retrieval augmented generation (RAG), semantic search or modern AI applications such as chatbots and knowledge bases has probably come across embedding models and rerankers. These models are essential for converting texts into vectorized representations that machines can efficiently compare and retrieve relevant information.
Today we introduce you to two new Managed AI models in our portfolio.
BAAI bge-m3 and bge-reranker-v2-m3 are live!
- BAAI/bge-m3 is currently the most popular multilingual embedding model
- BAAI/bge-reranker-v2-m3 offers a very strong price-performance ratio when reranking
bge-m3: The Swiss army knife among embeddings
Why so many people are currently so enthusiastic about it:
- Supports over 100 languages at a very high level (incl. German, French, Spanish, Arabic, Chinese, Japanese, …)
- Context length up to 8192 tokens, perfect for long documents, manuals, legal texts, research papers
- Supports dense retrieval, multi-vector retrieval and sparse retrieval
Typical use cases:
- Semantic & hybrid search
- RAG of any size
- Mehrsprachige Wissensdatenbanken
bge-reranker-v2-m3: The quality booster for your top-k results
A good retriever alone is often no longer enough. The Reranker sorts your top 20 or top 50 candidates much more precisely and this is exactly where this model shines.
Why is the model so popular?
- State-of-the-art in multilingual reranking
- Significantly more precise than bi-encoder alone
- Fast processing
Typical use cases:
- Two-stage retrieval
- Significantly improving RAG quality
- Question-answer systems, support & internal search
Integration made easy
The models are OpenAI-compatible. This means that you can use them almost anywhere where you already use embeddings or reranking or would like to use them in the future.
Well-known frameworks & tools that work out of the box:
- LangChain / LlamaIndex
- Haystack
- n8n
- Open-WebUI
- LM Studio
Quickstart with NWS Managed AI Models
You can integrate our newest models into your existing AI workflows in just 3 steps.
Step 1: Configure the API URL
The base URL for the API endpoints of both models is https://api.ai.nws.netways.de. The required slugs for both models are then as follows:
- For embedding:
/v1/embeddings - For reranking:
/rerank
Step 2: Specify model ID
For communication via the OpenAI-compatible API endpoints, you must specify the ID of the desired model in addition to the API URL. These are as follows:
- Embeddings:
BAAI/bge-m3 - Reranking:
BAAI/bge-reranker-v2-m3
Step 3: Store the API key and get started!
Now you just need to generate an API key in MyNWS or configure an existing key. You can already use our latest Managed AI models.
With bge-m3 and bge-reranker-v2-m3 you have one of the currently most powerful, flexible and at the same time most affordable open source retrieval stacks at your fingertips, without having to worry about model hosting, quantization or GPU availability.
Whether you are building a multilingual chatbot, an internal search or a really good RAG system: this combination will deliver state-of-the-art quality with manageable effort in 2026.
Have fun trying them out!
0 Comments