# BGE M3

Multilingual reranker that scores query-document relevance with an 8K-token input window. Input longer than the context length is silently truncated by the engine and billed as truncated, not rejected.

Model id: `bge-reranker-v2-m3`
Context: 8K tokens
Quantization: fp16
Regions: SJC
Hugging Face: [BAAI/bge-reranker-v2-m3](https://huggingface.co/BAAI/bge-reranker-v2-m3)

## Pricing

- Tokens: $0.05 / 1M tokens

## Inputs

- text: context 8K tokens

## Output: rerank

| Parameter | Type | Values | Default |
| --- | --- | --- | --- |
| `top_n` | integer | 1–100 |  |
| `return_documents` | boolean |  |  |

## Endpoints

- [`POST /v1/rerank`](/docs/reference/create-rerank.md): Rerank Documents