Implementation:Cohere ai Cohere python V2Client Rerank

Field	Value
Type	Implementation
Source	Cohere Python SDK
Domain	NLP Information Retrieval Reranking
Last Updated	2026-02-15
Implements	Principle:Cohere_ai_Cohere_python_Semantic_Reranking

Overview

Concrete method for reranking documents by semantic relevance using Cohere's cross-encoder models.

Description

V2Client.rerank takes a query string and list of document strings, sends them to the Cohere rerank API, and returns a V2RerankResponse with documents ordered by relevance score. Supports top_n filtering and max_tokens_per_doc truncation. Long documents are automatically truncated to max_tokens_per_doc (default 4096).

Code Reference

src/cohere/v2/client.py Lines L492-569

Signature

def rerank(
    self, *, model: str, query: str, documents: typing.Sequence[str],
    top_n: typing.Optional[int] = OMIT, max_tokens_per_doc: typing.Optional[int] = OMIT,
    priority: typing.Optional[int] = OMIT, request_options: typing.Optional[RequestOptions] = None,
) -> V2RerankResponse:

Import

from cohere import ClientV2 (access via client.rerank())

Inputs

Parameter	Type	Required	Description
model	str	Yes	The rerank model to use
query	str	Yes	The query string to rank against
documents	Sequence[str]	Yes	List of candidate document strings
top_n	Optional[int]	No	Number of top results to return
max_tokens_per_doc	Optional[int]	No	Max tokens per document (default 4096)

Outputs

V2RerankResponse with results list containing index, relevance_score (0-1).

Example

from cohere import ClientV2
client = ClientV2()
response = client.rerank(
    model="rerank-v4.0-pro",
    query="What is the capital of the United States?",
    documents=["Washington D.C. is the capital.", "Paris is in France.", "London is in England."],
    top_n=2,
)
for result in response.results:
    print(f"Document {result.index}: score={result.relevance_score:.4f}")

Related

• Principle:Cohere_ai_Cohere_python_Semantic_Reranking