Nemotron 3 Embed 1B
Awaiting activationRetrieve multilingual passages for business question-answering with NVIDIA’s newer text model.
What it does
Nemotron 3 Embed 1B is NVIDIA’s newer text retrieval model, trained for multilingual and cross-language similarity. NVIDIA positions it as a foundation for question-answer applications over large text collections and marks it ready for commercial use.
It produces 2,048-dimensional vectors and distinguishes queries from indexed passages. The full model card lists 32,768 tokens, while the hosted API reference limits inputs to 4,096 tokens; this integration follows that hosted contract and a 4,000-byte cap, with truncation disabled.
Language coverage
NVIDIA reports evaluation across 34 languages, including Arabic, Chinese, Japanese, Korean, Hindi, Bengali, Swahili and Vietnamese. Evaluation coverage does not guarantee identical quality across dialects or language pairs.
Official language coverage ↗Where it helps
International knowledge
Retrieve relevant text across supported languages.
Question-answer retrieval
Find source passages before a business assistant writes an answer.
Technical archives
Evaluate retrieval over a multilingual collection of manuals and reference documents.
When to choose it
A candidate for a multilingual enterprise retrieval project or a NVIDIA model evaluation. Compare results with Voyage, Cohere and Qwen on the same customer questions.
Awaiting a commercial serving arrangement, credentials and verified rates. Hosted evaluation access is not an unlimited production entitlement; images and other multimodal inputs are outside this text endpoint.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
NVIDIA · Nemotron 3 Embed 1B model card ↗Reviewed October 4, 2026