Rerank 5
A reranking model from Cohere built on a cross-encoder architecture, used for fine-grained ordering of candidate documents in RAG pipelines, available in Pro (accuracy-focused) and Fast (speed-focused) versions.
Timeline (last 90 days)
The core mechanism of Rerank 5 is Context Focus: it first locates the passages in a document relevant to the query, then scores relevance based on those passages, rather than compressing and scoring the entire text.
Rerank 5 provides calibrated relevance scores, making scores comparable across queries so that developers can set universal thresholds for filtering.
Rerank 5 is priced at $0.05 per million tokens.
The article points out that the latency and cost trade-offs brought by the 32K context still need to be verified in actual pipelines, and that public comparative data on the accuracy gap between Pro and Fast is also lacking.
For teams building or optimizing RAG systems, Rerank 5 is worth including in candidate evaluations, but the final choice should be based on measured results from their own document formats and query distributions.
Cohere released the reranking model Rerank 5, including Pro and Fast versions.
Currently, information about Rerank 5 mainly comes from Cohere's official release, and no third-party independent evaluation has yet appeared.
Cohere claims that Rerank 5's performance far surpasses Rerank 4 and all other models they tested.
Rerank 5 introduces a new feature called Context Focus, which can extract only the specific passages in a document relevant to the query to reduce token consumption, and it is optional to enable with an adjustable amount of retained context.
Rerank 5 is already available as an integration through LangChain and has launched on the Cohere API and several other platforms.
1 more timeline events
All Facts (11)
The article points out that the latency and cost trade-offs brought by the 32K context still need to be verified in actual pipelines, and that public comparative data on the accuracy gap between Pro and Fast is also lacking.
50%UnverifiedThe core mechanism of Rerank 5 is Context Focus: it first locates the passages in a document relevant to the query, then scores relevance based on those passages, rather than compressing and scoring the entire text.
50%UnverifiedRerank 5 provides calibrated relevance scores, making scores comparable across queries so that developers can set universal thresholds for filtering.
50%UnverifiedRerank 5 is priced at $0.05 per million tokens.
50%UnverifiedFor teams building or optimizing RAG systems, Rerank 5 is worth including in candidate evaluations, but the final choice should be based on measured results from their own document formats and query distributions.
50%UnverifiedCurrently, information about Rerank 5 mainly comes from Cohere's official release, and no third-party independent evaluation has yet appeared.
50%UnverifiedCohere claims that Rerank 5's performance far surpasses Rerank 4 and all other models they tested.
50%UnverifiedRerank 5 introduces a new feature called Context Focus, which can extract only the specific passages in a document relevant to the query to reduce token consumption, and it is optional to enable with an adjustable amount of retained context.
50%UnverifiedRerank 4 is Cohere's previous-generation product, with the Pro version emphasizing accuracy and the Fast version emphasizing throughput and latency, and this dual-version strategy continues into Rerank 5.
50%UnverifiedCohere released the reranking model Rerank 5, including Pro and Fast versions.
50%UnverifiedRerank 5 is already available as an integration through LangChain and has launched on the Cohere API and several other platforms.
50%