Overview
About this model
bge-reranker-v2-m3，Different from embedding model, reranker uses question and document as input and directly output similarity instead of embedding. You can get a relevance score by inputting query and passage to the reranker. And the score can be mapped to a float value in [0,1] by sigmoid function.
View model card on HuggingFaceBGE Reranker v2 M3 API for multilingual RAG
Rerank retrieved passages with BGE Reranker v2 M3 on EcoHash. Review multilingual RAG use cases, API examples and current token pricing.
When to choose it
- Reordering a shortlist of search results by relevance to a question before passing context to an LLM.
- Evaluating retrieval quality for multilingual documentation and knowledge bases.
Before you integrate
- A reranker scores query–passage pairs; it does not generate embeddings or search your database. Retrieve candidates first.
- Call POST /v1/rerank with a query and documents. This endpoint uses HTTP requests rather than an OpenAI SDK method.
- Evaluate relevance and latency on your own queries before changing candidate counts or the number of passages passed downstream.
Specifications
Model details
- Provider
- BAAI
- Modality
- Reranker
- Pricing
- $0.03 / 1M
Modalities: text → text
API
Call it in code
Use the EcoHash HTTP endpoint shown below with your API key. This endpoint has no matching OpenAI SDK method.
import requests
BASE_URL = "https://api.ecohash.com/v1"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}
ranked = requests.post(
f"{BASE_URL}/rerank",
headers=HEADERS,
json={
"model": "bge-reranker-v2-m3",
"query": "How is GPU runtime billed?",
"documents": [
"GPU runtime is billed per second at the current GPU-hour rate.",
"Persistent storage is billed separately until deleted.",
"SSH connects a terminal to your running workspace.",
],
"top_n": 2,
},
).json()curl https://api.ecohash.com/v1/rerank \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d {
"model": "bge-reranker-v2-m3",
"query": "How is GPU runtime billed?",
"documents": [
"GPU runtime is billed per second at the current GPU-hour rate.",
"Persistent storage is billed separately until deleted.",
"SSH connects a terminal to your running workspace."
],
"top_n": 2
}FAQ
Common questions
- How much does the BGE-Reranker-V2-M3 API cost?
- BGE-Reranker-V2-M3 is $0.03 per 1M tokens on EcoHash. That figure comes live from the EcoHash billing system, so the price shown on this page is the price you are billed — there is no separate list price to reconcile.
- What are the BGE-Reranker-V2-M3 limits on EcoHash?
- BGE-Reranker-V2-M3 is served with text → text modalities. The specifications table on this page is generated from the same API the runtime uses, so it cannot drift from what the endpoint accepts.
- Can I use BGE-Reranker-V2-M3 commercially through EcoHash?
- EcoHash hosts BGE-Reranker-V2-M3 under the model publisher's own licence and does not change its terms, so whether your use is permitted is decided by that licence — read it on the model card at https://huggingface.co/BAAI/bge-reranker-v2-m3. The EcoHash Terms of Service additionally require you to comply with each model's licence and acceptable-use policy. Your EcoHash bill for API usage is separate from, and does not grant, any right under the model's licence.
Guides and next steps
- Build a retrieval and reranking pipeline →
Use embeddings to retrieve candidates, then rerank the passages supplied to your model.
- GPU and model API pricing →
Check current rates before choosing a workspace, dedicated endpoint or shared API.