Cohere Rerank 4 Pro

Advanced cross-encoder reranking models (Fast & Pro) optimized for enterprise search and RAG, featuring a 32K context window, strong performance on long and complex documents, and multilingual retrieval across 100+ languages. Built for high-stakes domains like finance, healthcare, manufacturing, and e-commerce, with self-learning to adapt to domain-specific data over time. If you want to compare the best rerankers for your data, try Agentset.

Leaderboard Rank

#2 of 12

ELO Rating

1629

#2

Win Rate

57.7%

#2

Accuracy (nDCG@10)

0.095

#5

Latency

614ms

#7

Model Information

  • Provider: Cohere
  • License: Proprietary
  • Price per 1M tokens: $0.050
  • Release Date: 2025-12-11
  • Model Name: rerank-v4.0-pro
  • Total Evaluations: 3300

Performance Record

  • Wins: 1903 (57.7%)
  • Losses: 1270 (38.5%)
  • Ties: 127 (3.8%)

Rerankers Are Just One Piece of RAG

Agentset gives you a managed RAG pipeline with the top-ranked models and best practices baked in. No infrastructure to maintain, no reranking to configure.

Performance Overview

Cohere Rerank 4 Pro's ELO performance varies across different benchmark datasets, showing its strengths in specific domains.

Cohere Rerank 4 Pro - ELO by Dataset

  • arguana: ELO 1783 WR 364 W - 184 L - 2 T
  • FiQa: ELO 1702 WR 326 W - 207 L - 17 T
  • business reports: ELO 1658 WR 341 W - 197 L - 12 T
  • MSMARCO: ELO 1586 WR 286 W - 209 L - 55 T
  • PG: ELO 1553 WR 313 W - 237 L - 0 T
  • DBPedia: ELO 1491 WR 273 W - 236 L - 41 T

Dataset Breakdown

arguana

  • ELO: 1783
  • Win Rate: 66.2%
  • Accuracy Metrics:
    • nDCG@50: 0.353
    • nDCG@100: 0.439
    • Recall@50: 0.660
    • Recall@100: 0.920
  • Latency Distribution:
    • Mean: 785ms
    • P50 (Median): 768ms
    • P90: 933ms

FiQa

  • ELO: 1702
  • Win Rate: 59.3%
  • Accuracy Metrics:
    • nDCG@50: 0.126
    • nDCG@100: 0.129
    • Recall@50: 0.130
    • Recall@100: 0.135
  • Latency Distribution:
    • Mean: 610ms
    • P50 (Median): 585ms
    • P90: 817ms

business reports

  • ELO: 1658
  • Win Rate: 62.0%
  • Accuracy Metrics:
    • nDCG@50: 0.000
    • nDCG@100: 0.000
    • Recall@50: 0.000
    • Recall@100: 0.000
  • Latency Distribution:
    • Mean: 529ms
    • P50 (Median): 498ms
    • P90: 675ms

MSMARCO

  • ELO: 1586
  • Win Rate: 60.0%
  • Accuracy Metrics:
    • nDCG@50: 0.000
    • nDCG@100: 0.000
    • Recall@50: 0.000
    • Recall@100: 0.000
  • Latency Distribution:
    • Mean: 458ms
    • P50 (Median): 408ms
    • P90: 615ms

PG

  • ELO: 1553
  • Win Rate: 56.9%
  • Accuracy Metrics:
    • nDCG@50: 0.000
    • nDCG@100: 0.000
    • Recall@50: 0.000
    • Recall@100: 0.000
  • Latency Distribution:
    • Mean: 760ms
    • P50 (Median): 720ms
    • P90: 896ms

DBPedia

  • ELO: 1491
  • Win Rate: 38.6%
  • Accuracy Metrics:
    • nDCG@50: 0.000
    • nDCG@100: 0.000
    • Recall@50: 0.000
    • Recall@100: 0.000
  • Latency Distribution:
    • Mean: 541ms
    • P50 (Median): 489ms
    • P90: 729ms

Build RAG in Minutes, Not Months

Agentset gives you a complete RAG API with top-ranked rerankers and embedding models built in. Upload your data, call the API, and get accurate results from day one.

Example Code

import { Agentset } from "agentset";

const agentset = new Agentset();
const ns = agentset.namespace("ns_1234");

const results = await ns.search(
  "What is multi-head attention?"
);

for (const result of results) {
  console.log(result.text);
}

Compare Models

Compare Cohere Rerank 4 Pro with other top rerankers to understand the differences in performance, accuracy, and latency.