Back to Videos
BM25 vs OpenAI Embeddings: Why a 1994 Ranker Still Wins
81
Retrieval & Search Research
Mixpeek Team
July 14, 2026
Summary
A keyword ranking function from 1994 beat OpenAI's best embedding models on standard retrieval benchmarks. The video explains how BM25 works, where dense embeddings genuinely win, and why production search stacks run both stages rather than picking a side.
bm25embeddingsopenairankinghybrid-search
About this video
A keyword ranking function from 1994 beat OpenAI's best embedding models on standard retrieval benchmarks. The video explains how BM25 works, where dense embeddings genuinely win, and why production search stacks run both stages rather than picking a side.