GPT Researcher drops embeddings: LLM-based retrieval lifts relevant context 59% at same cost
hwchase17 · x · 2026-09-28
- GPT Researcher replaced embeddings with an LLM (Jev) in its RAG pipeline and benchmarked both on 28 research tasks from SimpleQA plus open-ended research.
- Results: 59% more relevant context (73% vs 46%), reports preferred 15 to 3 in blind comparisons, and identical cost per report. GPT Researcher now runs without embeddings by default.
- The team claims LangChain + Tavily + LLM is all you need for a full RAG setup; the repo and research are open-sourced.
- LangChain founder Harrison Chase amplified the result, calling it unexpected.
More from coding & agent
- OCR'd millions of pages of well data so AI agents can search old oil fields — oilmutt · 2026-09-28
- Google Antigravity adds native inter-agent messaging and user-to-agent DMs — rseroter · 2026-09-28
- Dataviz expert says AI pushes him toward lower-level charting languages for finer control — randal_olson · 2026-09-28
- Cloudflare Birthday Week 2026: native Rust in Workers, agentic browser, EmDash CMS — ritakozlov · 2026-09-28
- Creative arbitrage: build a Claude Code skill to ship 500+ static ads for almost nothing — alexgoughcooper · 2026-09-28
- 4B Decision Model Mica Mines a Diamond Pickaxe in 26 Steps on First Real-Minecraft Run — Top-Evidence174 · 2026-09-28