Cohere Rerank 3.5
The default commercial cross-encoder reranker for two-stage retrieval
About
Cohere Rerank 3.5 is a cross-encoder reranking model: instead of embedding queries and documents separately, it reads each query-document pair together and returns a relevance score, which typically lifts ranking quality well beyond raw vector search. It supports 100+ languages, English and multilingual endpoints, semi-structured data like JSON, and a 4096-token context with automatic chunking of longer documents. Launched December 2, 2024, it became the procurement default for enterprise RAG reranking, available on Cohere's API, AWS Bedrock, and SageMaker. The honest math: reranking adds latency (a few hundred ms per query) and per-search cost on top of your embedding bill, and Cohere's own numbers show the biggest gains on constrained queries and semi-structured data rather than generic prose.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Cohere-reported gains over Rerank v2 on constrained queries and semi-structured JSON
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Dec 2, 2024 launch verified in the AWS Bedrock model card; 100+ languages, 4K context, JSON support corroborated across Bedrock docs and integration writeups.; Pricing ($2.00/1K searches) is a third-party 2026 figure — labeled as such.; Latency/cost tradeoffs of the reranking stage stated honestly..