NVIDIA open-sources Nemotron 3.5 Embed, Rerank and Parse models to cut agent inference token waste
NVIDIAAI · x · 2026-10-30
NVIDIA expanded its Nemotron Retriever collection with open, commercially usable retrieval models: Nemotron 3.5 Embed (8B BF16, 1B BF16, 1B NVFP4), Nemotron 3.5 Rerank 8B, and the 631M-parameter Nemotron 3.5 Parse.
The pitch: every extra search an agent makes burns tokens, and weak retrieval makes agents re-search and ingest irrelevant documents. Better embedding, reranking, and document parsing let agents reach grounded answers with far less wasted inference cost at enterprise scale.
The models bring multimodal retrieval, relevance ranking, and document processing to agent workflows. NVIDIA published comparisons against another open-source retrieval stack on Hugging Face, covering retrieved evidence, accuracy, and token usage, plus open models and recipes.
Related event: NVIDIA Open-Sources Nemotron 3.5 Retriever Suite(2 posts)→
More from coding & agent
- Particle-1 matches DeepSeek V4.1 Flash with 43% fewer output tokens on Real-SWE — ycombinator · 2026-10-30
- He put all global law into Git with Claude Code — and hit #1 on HN — SnooPeripherals5313 · 2026-10-30
- Stanford & Northeastern unveil Retriever, a time-aware programming framework for long-horizon robots — yining_hong · 2026-10-30
- Debate: The Atomic Unit of Workflows Should Be the Decision, Not the Workflow — round · 2026-10-30
- Neuralink Prototype: Think "Hey Neuralink" to Trigger a Grok Computer-Use Agent — Baconbrix · 2026-10-30
- Founder betting all-in on a general autoprover: formal verification without LLM-generated proofs — CatAstro_Piyush · 2026-10-30