Wispr Flow - Don’t type, just speak
Mini toolsRerank 2.5 lite
By Voyage AI
Model family: Rerank
rerank-2.5-lite is a latency-optimized reranker from Voyage AI's rerank-2.5 series. It takes a query and a set of candidate documents and outputs relevance scores to re-order results for improved retrieval accuracy. The model supports a 32K token context window, double that of its predecessor rerank-2-lite and 8x that of Cohere Rerank v3.5, enabling reranking of longer documents without truncation. A key addition is instruction-following: users can append natural language instructions to their queries to define relevance criteria, specify document types, or disambiguate queries. Evaluated across 93 retrieval datasets spanning 9 domains and 31 languages, rerank-2.5-lite outperforms Cohere Rerank v3.5 by 7.16% and beats Qwen3-Reranker-8B despite being over an order of magnitude smaller. On the MAIR instruction-following benchmark, it outperforms Cohere Rerank v3.5 by 10.36%. Available via API with token-based pricing.
Text
Gen 7
Overview
A latency-optimized reranker that refines retrieval results with instruction-following capabilities. Supports a 32K token context window, works across 31 languages, and outperforms larger open-source rerankers. Users can steer relevance scoring using natural language instructions appended to queries.
About Voyage AI
Voyage AI provides best-in-class embedding models and rerankers for search and retrieval over unstructured data, used to power retrieval-augmented generation (RAG) and AI applications. It offers general-purpose, domain-specific (finance, legal, code) and company-specific fine-tuned models. Founded in 2023 and based in Palo Alto, the company was acquired by MongoDB, Inc. in February 2025 and now operates as a MongoDB subsidiary.
View Company Profile
Last updated: June 24, 2026
