I'm looking for a model that would optimize for accuracy (finding relevant results), language support (as many as possible, English only is a no-no) and efficiency so I could potentially run it locally or that there's a cheap API for it.
OpenAI embedding API gets expensive real quick when generating embeddings for 10^5 documents and more.