🔍 Read the full analysis: How NVIDIA Nemotron 3.5 Embed, Rerank, And Parse Advances AI Agents on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
NVIDIA has expanded its Nemotron Retriever collection with five models for embedding, reranking and document parsing. The company reports improved retrieval scores and lower inference-token use in selected tests, but the supplied announcement lacks independent validation and detailed information about the Parse model.
NVIDIA has expanded its Nemotron Retriever collection with five models for embedding, reranking and document parsing, targeting AI agents that need to retrieve evidence from text and image-rich business documents, as described in the original analysis. NVIDIA says selected evaluations showed better retrieval results alongside lower agent inference-token use, but the reported findings are company benchmarks, not independent validation.
The collection comprises Nemotron 3.5 Embed 8B BF16, Embed 1B BF16, Embed 1B NVFP4, Rerank 8B BF16 and Parse, with 631 million parameters. NVIDIA positions the embedding models for finding relevant material across text and visual content, including tables, charts, infographics, datasheets and schematics. The reranker scores query-document pairs to help an agent prioritize candidate evidence.
NVIDIA says it evaluated retrieval configurations on ViDoRe v3, BRIGHT and BrowseComp-Plus, holding the language model, tools, tasks, corpus and execution policy fixed. In those tests, the company reports that adding Nemotron 3.5 Rerank reduced agent inference tokens per query by about 14% with Embed 8B and 13% with Embed 1B, while improving retrieval accuracy. It also reports that Embed 8B scored about 0.9 points higher in nDCG@10 than Qwen3-VL-Reranker-8B at a similar agent token cost.
On ViDoRe v3.1, NVIDIA reports an nDCG@10 score of 61.17 for Embed 8B, compared with 52.23 for Llama Nemotron Embed VL 1B. Embed 1B BF16 scored 58.38, which NVIDIA describes as 95.4% of the 8B model’s score. These results reflect the company’s stated evaluation setup and do not establish how the models will perform on other data or workloads.
Retrieval Can Cut Agent Work
Retrieval affects both the evidence available to an agent and the amount of work it may need to do before answering. When a system repeatedly searches irrelevant material or misses information in a chart, it may need additional searches and processing. Better ranking could help an agent focus on more useful documents, while lowering repeated inference work for organizations running agents across large collections.
The models’ emphasis on visual and text-based evidence addresses a common document-search limitation: important information may appear in a chart, table or schematic rather than in searchable text. If the reported gains hold in a company’s own environment, multimodal retrieval could broaden the kinds of evidence agents can use. But outcomes will depend on the document collection, workload, deployment hardware and system integration; the benchmarks do not establish savings or accuracy gains for every use case.
As an affiliate, we earn on qualifying purchases.
Embedding, Ranking and Parsing
Embedding models represent queries and documents in a form that search systems can compare for semantic relevance. A reranker then scores candidate documents against a query, refining which items an agent sees first. NVIDIA describes an agent workflow in which a system searches, inspects evidence and may search again based on what it finds. The new models extend the company’s earlier Nemotron 3 Embed text embedding collection toward multimodal retrieval.
NVIDIA presents the 8B BF16 model as a quality-focused option and Embed 1B BF16 as a smaller multimodal alternative. Embed 1B NVFP4 applies quantization to text and image encoders for inference efficiency on supported hardware. The company also describes fine-tuning recipes for custom collections, reporting relative nDCG@10 gains in chart and infographic experiments of 18% for Embed 1B BF16 and 3% for Embed 8B BF16.
The announcement says text retrieval was evaluated across 34 languages, while multilingual visual retrieval was tested on MIRACL Vision across 18 languages and on KoViDoRe. It does not provide results for languages beyond the stated evaluation sets. The supplied material identifies Nemotron 3.5 Parse as a 631-million-parameter model but offers little information about its role or performance.
“Adding Nemotron 3.5 Rerank reduces agentic token cost per query by approximately 14% with Nemotron 3.5 Embed 8B, and 13% with Nemotron 3.5 Embed 1B, while improving retrieval accuracy in both configurations.”
— NVIDIA, in the Hugging Face announcement
As an affiliate, we earn on qualifying purchases.
Benchmark and Parse Gaps
The supplied announcement does not include full benchmark methodology, complete results tables or independent replication of the reported scores and token savings. It also does not specify the publication date. Performance on other corpora, languages and agent tasks remains unestablished, and the figures should be read as results from NVIDIA’s stated evaluations rather than a general performance guarantee.
Details about Nemotron 3.5 Parse are especially limited: the source gives its parameter count but not its supported formats, capabilities, evaluation results or how it fits into the embedding and reranking workflow. The source describes the collection as open and commercially available but does not provide licensing terms, pricing, release links, full hardware requirements or a broader availability schedule.
visual and text evidence analysis tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Documentation and Independent Tests
Users seeking to evaluate the models will need the model artifacts and documentation to check licensing, deployment requirements and reproducibility. Independent tests on enterprise document collections could show whether NVIDIA’s reported retrieval improvements and token reductions carry over to different data and workflows. Further documentation on Parse’s capabilities and evaluations would clarify what it adds to the collection. The supplied source does not give dates for those materials or further updates.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did NVIDIA announce?
NVIDIA expanded its Nemotron Retriever collection with five models for embedding, reranking and document parsing, including three embedding variants, one reranker and the 631-million-parameter Parse model.
What performance gains did NVIDIA report?
NVIDIA reports improved retrieval accuracy and about 14% fewer agent inference tokens per query with Embed 8B, and about 13% fewer with Embed 1B, when adding its Rerank model in selected tests. These are company-reported results, not a guarantee for other workloads.
What content can the embedding models retrieve?
NVIDIA describes them as supporting retrieval across text and visual material, including tables, charts, infographics, datasheets and schematics. The results depend on the documents and deployment setup.
What is known about Nemotron 3.5 Parse?
The supplied source identifies Parse as a 631-million-parameter model, but does not give full details about its supported formats, capabilities, availability or evaluation results.
Have the benchmark results been independently verified?
The supplied announcement provides NVIDIA’s reported evaluations but no independent replication. Results on other datasets, languages and agent tasks remain unclear.
Primary source: Hugging Face · via ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
