Study: LLMs Marginally Beat Embedding Models but at 1000x Cost

A new paper comparing 10 LLMs against 26 embedding models across 37 tasks found LLMs scored only marginally higher (77.6 vs 77.2) at roughly 1000x the cost. The authors recommend LLMs mainly for reasoning-heavy retrieval, and embedding models for classification and semantic similarity.

2026-08-20 ~ 2026-08-20 · 2 related posts

• Paper: LLMs Beat Embeddings Slightly but Cost 1,431x More — Muennighoff · 2026-08-20
• Author's guidance: LLM embeddings only worth it for reasoning-heavy retrieval — Muennighoff · 2026-08-20