The Embedder's Dilemma: LLMs Are Better, but at What Cost?

• 类型：arxiv
• 标识：2608.12875
• 链接：https://arxiv.org/abs/2608.12875
• 主分类：rag
• 形态：method
• TLDR：Should you replace your text-embedding pipeline with a large language model? We answer this with a controlled, cost-aware comparison of ten LLMs across six families and 26 embedding models (118M to 14B parameters) on 37 tasks spanning classification, semantic textual similarity (STS), clustering, pair classification, and retrieval. In aggregate the two paradigms are effectively tied: the best LLM (Gemini 3.1 Pro, 77.6) and the best embedding model (77.2) differ by 0.4 points. Their strengths differ by task: LLMs lead on reasoning-heavy retrieval, embedding models lead on classification, and th
• 待LLM分类：否
• 标题中文：嵌入器的困境：LLM 更强，但代价几何？
• TLDR中文：是否应将文本嵌入流水线替换为 LLM？我们对此在 37 个任务（涵盖分类、语义文本相似度 STS、聚类、配对分类与检索）上，对覆盖 6 个家族、参数规模 118M 至 14B 的 10 个 LLM 与 26 个嵌入模型进行了受控且考虑成本的对比。总体上两种范式基本持平：最佳 LLM（Gemini 3.1 Pro，77.6）与最佳嵌入模型（77.2）仅相差 0.4 分。两者在不同任务上各有所长：LLM 在推理密集型检索上领先，嵌入模型则在分类任务上领先，且两者……
• 来源文件：
• /inbox/tom/_candidates/2026-08-22-agent-rag-longcontext-candidates.json
• /inbox/tom/_candidates/2026-08-23-agent-rag-longcontext-candidates.json