OpenLuxEmbeddings & RAG
A retrieval pipeline has three cost centres — embedding, reranking and answer generation. On OpenLux all three run on one key: BGE-M3 embeds, Jina Reranker v3 reorders, and a budget text tier writes the answer.
Retrieval-augmented generation fails on retrieval quality far more often than on generation quality. That is why the pipeline below spends its first two stages on search: multilingual embeddings from BGE-M3 go into your vector database, and the reranker rescores the top candidates before the answer model ever sees them.
Embeddings are standard OpenAI-shaped vectors on /v1/embeddings, so any vector database or framework that speaks the OpenAI API works without an adapter. The whole pipeline is metered per 1M tokens in USD, and with rates this low the embedding and reranking stages of thousands of queries cost cents.
The picks
4 models from the live catalogue, with the rate each one is metered at.
Jina Reranker V3
Reranking that lifts retrieval accuracy before generation — $0.0042 per 1M tokens.
Gemini 3 Flash (preview)
Cheapest generation tier for grounded answers at volume — $0.0368 per 1M tokens (-92% vs official rates).
GPT-5.6 Sol
Stronger generation when answer quality drives the product — $0.1838 per 1M tokens (-96% vs official rates).
Embeddings & RAG questions
Which retrieval models does OpenLux offer?
BGE-M3 for multilingual embeddings and Jina Reranker v3 for reranking, both metered per 1M tokens at $0.0035 and $0.0042 respectively.
How do I build a RAG pipeline with OpenLux?
Embed documents with BGE-M3 through the OpenAI-compatible embeddings endpoint, store the vectors in your database of choice, retrieve candidates, rerank them with Jina Reranker v3, then generate the answer with a low-cost text model. One key covers every stage.
What does a RAG query cost end to end?
Embedding and reranking cost fractions of a cent per query at the rates above, and generation from $0.0368 per 1M tokens on the flash tier. Thousands of grounded queries per month typically cost less than a few dollars in total.
Do OpenLux embeddings work with my existing vector database?
Yes. They are standard OpenAI-shaped embedding vectors, so any vector store or framework with an OpenAI-compatible client — from pgvector to Pinecone and LangChain — works by pointing the base URL at OpenLux.
Compare the picks head-to-head
BGE-M3 vs Jina Reranker V3
Rates, strengths and which model to pick.
GPT-5.6 Sol vs Gemini 3 Flash (preview)
Rates, strengths and which model to pick.
Gemini 3 Flash (preview) vs Claude Sonnet 4.5
Rates, strengths and which model to pick.
Gemini 3 Flash (preview) vs Claude Opus 5
Rates, strengths and which model to pick.
Gemini 3 Flash (preview) vs Claude Fable 5.1
Rates, strengths and which model to pick.
Gemini 3 Flash (preview) vs Gemini 3.1 Pro (preview)
Rates, strengths and which model to pick.
Put these picks behind one key
OpenLux meters every model in this collection on the same OpenAI-compatible endpoint, with pay-as-you-go billing and free trial credit to start.