RELEASE · MODELS · #2205
Nvidia unveils Nemotron 3.5 Super VL and lightweight Bolt models for enterprise agents and local AI PCs
Nvidia announced new open-weight models aimed at always-on agents and edge/PC deployment: Nemotron 3.5 Super VL (120B parameters) for enterprise agents, plus two lightweight Bolt models including Nemotron 3.5 Bolt 74B-A9B 24GB and the smaller Bolt E4B for local AI PCs and workstations. Nvidia says Super VL and Bolt 74B-A9B are available now (Bolt E4B coming later this month), claims Super VL can match larger Nemotron 3 Ultra on instruction-following and outperform some competing models in post-trained agent tasks, and named early users such as Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce.
KEY POINTS
- Nvidia announced new open-weight models aimed at always-on agents and edge/PC deployment: Nemotron 3.5 Super VL (120B parameters) for enterprise agents, plus two lightweight Bolt models including Nemotron 3.5 Bolt 74B-A9B 24GB and the smaller Bolt E4B for local AI PCs and workstations.
- Nvidia says Super VL and Bolt 74B-A9B are available now (Bolt E4B coming later this month), claims Super VL can match larger Nemotron 3 Ultra on instruction-following and outperform some competing models in post-trained agent tasks, and named early users such as Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce.
- This expands Nvidia's open-model lineup for on-premises and local agent deployment, enabling enterprises and PC users to run agentic models under their control and supporting Nvidia's push to sell AI‑capable hardware for edge and desktop use.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Nvidia announced new open-weight models for agent and edge/PC deployment: Nemotron 3.5 Super VL (120B parameters) for enterprise agents; two lightweight Bolt variants including Nemotron 3.5 Bolt 74B‑A9B (24 GB) and the smaller Bolt E4B for local AI PCs and workstations; and new Nemotron 3.5 retrieval models (Embed, Rerank, Parse). Nvidia says Super VL and Bolt 74B‑A9B are available now and Bolt E4B is coming later this month.
WHY NOW
This expands Nvidia’s open‑model lineup targeted at always‑on agents and local/edge deployment, aligning with enterprises’ preference for on‑premise/post‑training control and Nvidia’s push to place AI workload capability on PCs and edge hardware. The new retrieval models are presented as a way to reduce agent inference waste by improving evidence retrieval and ranking.
WHO IS AFFECTED
Directly affected parties named in the sources: enterprises building always‑on agents and doing post‑training; developers and knowledge workers on AI‑capable PCs and workstations; hardware/PC partners referenced (e.g., Microsoft and Dell in the media excerpt); and companies identified as early users (Skydance, EY, Ramp, CrowdStrike, Perplexity, Salesforce).
CONFIRMED
All facts below are directly from the supplied excerpts: - Nvidia announced Nemotron 3.5 Super VL at 120 billion parameters. (TechCrunch) - Nvidia announced Nemotron 3.5 Bolt 74B‑A9B (24 GB) and Bolt E4B (smaller). (TechCrunch summary & TechCrunch) - Nvidia stated Nemotron 3.5 Super VL and Bolt 74B‑A9B are available now; Bolt E4B is coming later this month. (summary & TechCrunch) - Nvidia claims Super VL can deliver similar instruction‑following and tool‑use accuracy to Nemotron 3 Ultra and, when post‑trained on agentic tasks, completes tasks faster than some competing models (examples listed in the excerpt). (TechCrunch) - Nvidia named early users: Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce. (TechCrunch) - Nvidia expanded Nemotron Retriever models: Nemotron 3.5 Embed (8B BF16, 1B BF16, 1B NVFP4), Nemotron 3.5 Rerank (8B BF16), and Nemotron 3.5 Parse (631M). The Hugging Face post states these bring multimodal retrieval, reranking, and document processing to enterprise agent workflows and that retrieval quality affects agent inference cost. (Hugging Face) - TechCrunch reports the Bolt models are intended to run locally on new AI PCs using Nvidia RTX Spark and on workstations running DGX Spark; and that such PCs are being released into the market through partners like Microsoft and Dell. (TechCrunch)
UNCERTAIN
Missing or unclear in the supplied excerpts: - Independent third‑party benchmark data verifying Nvidia’s performance claims versus Nemotron 3 Ultra or competing models is not provided in the excerpts. The Hugging Face post describes retrieval evaluations but the excerpt does not provide numeric results or full evaluation tables. - Scope and scale of the named customers’ deployments (pilot, production, extent of usage) are not specified. - Precise hardware requirements, memory/latency profiles, and platform compatibility details for running Super VL or Bolt models are not given in the excerpts beyond mention of RTX Spark and DGX Spark in the TechCrunch excerpt. - Exact release date for Bolt E4B is unspecified (only “later this month” in the excerpt). - Any licensing, redistribution terms, or commercial restrictions for the announced models are not detailed in the provided text. - Potential differences between Nvidia’s claimed test tasks/datasets and independent real‑world workloads are not documented in the excerpts. (Where evidence is missing or only asserted by Nvidia/TechCrunch, I label it as uncertain.)
WHAT TO WATCH
Concrete observable signals to monitor next (from supplied sources): - Official release or repository pages (Nvidia/Hugging Face) publishing model weights, checkpoints, and technical specs for Super VL, Bolt 74B‑A9B, Bolt E4B, and the Retriever models. - Published benchmarks, evaluation tables, or whitepapers comparing Super VL to Nemotron 3 Ultra and named competitor models on instruction‑following, tool use, and agentic tasks. - Announcement or product pages that confirm the Bolt E4B launch date and exact availability. - Customer case studies, press releases, or technical writeups from the named early users (Skydance, EY, Ramp, CrowdStrike, Perplexity, Salesforce) describing deployment scope and results. - Availability notices from PC partners (e.g., Microsoft, Dell) or product listings for AI PCs using RTX Spark / DGX Spark referenced in the media excerpt. - Release of detailed retrieval evaluation results or full ViDoRe/BRIGHT/BrowseComp‑Plus scores referenced in the Hugging Face post.
WHY IT MATTERS
This expands Nvidia's open-model lineup for on-premises and local agent deployment, enabling enterprises and PC users to run agentic models under their control and supporting Nvidia's push to sell AI‑capable hardware for edge and desktop use.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
Nvidia announced Nemotron 3.5 Super VL, a 120-billion-parameter model intended for enterprise agents.
SUPPORTEDChecked 2026-10-21 · 1 supporting
Nvidia launched two lightweight Bolt models, including Nemotron 3.5 Bolt 74B-A9B (24GB) and a smaller Bolt E4B targeted at local AI PCs and workstations.
SUPPORTEDChecked 2026-10-21 · 1 supporting
Nvidia said Nemotron 3.5 Super VL and Bolt 74B-A9B are available now, while Bolt E4B will be released later this month.
SUPPORTEDChecked 2026-10-21 · 1 supporting
Nvidia claims Super VL can match the larger Nemotron 3 Ultra on instruction-following and outperform some competing models after post-training on agentic tasks.
SUPPORTEDChecked 2026-10-21 · 1 supporting
SOURCES & TIMELINE
2An agent’s answers depend on the evidence it can retrieve. When that evidence is incomplete or irrelevant, the agent will repeat searches and inspect additional documents, consuming more inference tokens before completing the task. Better retrieval can help agents reach grounded answers with fewer wasted tokens. At enterprise scale, that waste leads to massive inefficiencies. Agents need the right retrieval models an…
Nvidia on Thursday launched new open-weight models designed for always-on agents and edge computing. The Nemotron 3.5 Super VL is a large, 120-billion-parameter model geared for agents used by the enterprise. While it uses fewer parameters than the Nemotron 3 Ultra model, Nvidia says it can deliver similar accuracy in instruction-following and tool-use tasks. At the other end of the spectrum, the company also launc…