ettin-reranker-1b-v1

by cross-encoder

State-of-the-art cross-encoder reranker matching its 1.5B teacher at 1B parameters

52Kdl/month

1Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

cross-encoder/ettin-reranker-1b-v1

Feature URI

mixpeek://reranker@v1/cross_encoder_ettin_1b_v1

Overview

Ettin Reranker 1B is the flagship model in the Ettin reranker family (17M to 1B parameters), trained via pointwise MSE distillation from the mxbai-rerank-large-v2 teacher. Built on Ettin ModernBERT encoders, it matches the teacher's MTEB Retrieval score within 0.0001 nDCG@10 while being smaller and faster.

The Ettin family provides a reranker at every size class, letting you trade latency for quality. The 150M variant runs under 10ms per query-document pair on GPU; the 1B variant delivers maximum accuracy for quality-critical retrieval.

Architecture

ModernBERT encoder backbone (Ettin variant) with a cross-encoder classification head. Takes concatenated query-document input and outputs a relevance score. Trained via pointwise MSE distillation from mxbai-rerank-large-v2 on diverse retrieval datasets.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

// Managed: create a collection over a bucket; Mixpeek runs this model's extractor
const collection = await mx.collections.create({
  namespace_id: "my-namespace",
  collection_name: "my-collection",
  source: { type: "bucket", bucket_ids: ["bkt_your_bucket"] },
  feature_extractor: {
    feature_extractor_name: "text_embedding",
    version: "v1",
    parameters: { model_id: "BAAI/bge-m3" },
  },
});