You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.

This is a reranker: it receives a search query and a list of documents, then re-evaluates their relevance and returns scores for re-ranking. It is suitable as a second stage in RAG and semantic search after retrieval from a vector database like Qdrant or pgvector.

The model does not write text: the output provides numerical relevance scores for the submitted query-document pairs. Plan for the input volume, not generation length — the context is 4095 tokens, meaning the query along with the document fragment must fit within it.

A separate reasoning mode is not noted in our data, and it is not needed for a reranker: the model compares the query with the document in a single pass and provides a score. If you need reasoning chains, connect a text model in BotHub — the API remains the same.

Function calling is not noted in our data. However, the reranker's response is machine-readable on its own: you receive relevance scores for each document and can easily parse them in your code to re-rank search results before passing them to a generative model.

The acceptance of other modalities is not noted in our data — work with text query-document pairs. If your pipeline requires images, audio, or video, switch to a suitable model in BotHub: the unified OpenAI-compatible API allows you to do this without rewriting the integration.

The M3 family from BAAI was originally designed as multilingual, so the reranker is also used for Russian-language collections. Quality still depends on your data, so measure ranking metrics on your own sample — through BotHub, this is available without a VPN or foreign card.