NVIDIA releases open Nemotron 3.5 Embed, Rerank and Parse models to cut agent token costs
NVIDIAAI · x · 2026-10-30
NVIDIA expanded its Nemotron Retriever collection with open, commercially usable models: Embed 8B/1B (incl. NVFP4), Rerank 8B, and a 631M-parameter Parse document-extraction model, covering multimodal retrieval, relevance ranking and document processing for enterprise agents.
The pitch: retrieval quality directly drives agent inference cost—irrelevant evidence makes agents repeat searches and read more documents, burning tokens. NVIDIA published head-to-head comparisons against another open retrieval stack showing retrieved evidence, accuracy and token usage; models and recipes are on Hugging Face.
Related event: NVIDIA Open-Sources Nemotron 3.5 Retriever Suite(2 posts)→
More from coding & agent
- Claude Code v2.1.314 adds write-blocking outside workdirs, idle compaction, gateway fixes — ashwin-ant · 2026-10-30
- Particle-1, a $0.25/$1 codegen model with 1M context, goes live on agent harness Capy — garrytan · 2026-10-30
- Aware demo: turning browser tabs into your own information universe with agents — HaijunXia · 2026-10-30
- Ex-Brex eng director runs Claude Code agents like a hundreds-person engineering org — garrytan · 2026-10-30
- Memgraph 3.14 ships 11 breaking changes and tighter null/NaN semantics — JeremyCMorgan · 2026-10-30
- ASC CLI 6.0: macOS publishing now requires PKG, --ipa with MAC_OS rejected — rudrank · 2026-10-30