Cohere¶

Reranking via Cohere's API.

Overview¶

Field	Value
Type	API
Provider	Cohere
Default Model	`rerank-v3.5`
Env Variable	`COHERE_API_KEY`

Installation¶

pip install "autorag-research[reranker]"
# or
uv add "autorag-research[reranker]"

Configuration¶

_target_: autorag_research.rerankers.cohere.CohereReranker
model_name: rerank-v3.5

Options¶

Option	Type	Default	Description
model_name	str	`rerank-v3.5`	Cohere model name
api_key	str	None	API key (or use env var)
batch_size	int	64	Batch size for multiple queries

Models¶

Model	Description
rerank-v3.5	Latest, best quality
rerank-english-v3.0	English optimized
rerank-multilingual-v3.0	Multilingual

Usage¶

from autorag_research.rerankers import CohereReranker

reranker = CohereReranker(model_name="rerank-v3.5")
results = reranker.rerank("What is RAG?", ["doc1", "doc2", "doc3"], top_k=2)

for r in results:
    print(f"[{r.index}] {r.score:.3f}: {r.text[:50]}...")