REFRAG speeds up RAG with reusable embeddings
A paper published on September 15, 2025 presents REFRAG, a method that replaces most retrieved tokens with precomputed chunk embeddings. It reports 30× higher speed and support for 16× longer contexts without loss of accuracy.
Source: REFRAG substitui tokens recuperados por embeddings reutilizáveis de chunks (x.com). Text prepared with AI from this source.
What happened and what to do
Published on September 15, 2025, the paper presents REFRAG, a method that replaces most tokens retrieved in RAG systems with precomputed, reusable chunk embeddings. The paper reports 30× higher speed and support for 16× longer contexts without loss of accuracy.
A company evaluating RAG can measure the cost and latency of its current workflow and test embedding reuse on a representative sample of queries and documents. Implementation could include a chunk preparation and update pipeline, quality and latency monitoring, and a comparison with the existing approach before broader adoption.
How the consultancy can help
Wendelmaques can diagnose the opportunity in a company’s RAG workflow, define an evaluation with metrics and quality criteria, and propose a scoped implementation for preparing, reusing, and monitoring embeddings.
Next step
Send a short description of your RAG use case to receive a scoped proposal for diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.