BGE Reranker v2-M3

BGE Reranker v2-M3 is the canonical second-stage reranker for RAG pipelines, pairs naturally with BGE-M3 embeddings (same family, same training corpus) but works with any first-stage retriever. Takes a query plus a candidate document set and re-orders by semantic relevance with much higher precision than vector cosine alone. 8192-token context per document. Standard pattern: vector recall returns top-50, reranker scores all 50, you keep top-5 for the LLM context.

Quick start

Use any OpenAI-compatible client. Just change base_url and your key.

from openai import OpenAI

client = OpenAI(
    api_key="tl-xxxxxxxxxxxxxxxxxxxxxxxx",
    base_url="https://api.thalam.ai/v1",
)

response = client.chat.completions.create(
    model="baai/bge-reranker-v2-m3",
    messages=[
        {"role": "user", "content": "Hello from Thalam!"}
    ],
)

print(response.choices[0].message.content)