← Back
NVIDIA Blog

Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models

•6 min read•
#agents#llm#nvidia
Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models
Level:Advanced
For:AI Engineers
✦TL;DR

NVIDIA unveiled the Nemotron 3.5 Super VL, a 120‑billion‑parameter open model that delivers frontier‑level multimodal agentic accuracy at roughly one‑fifth the size of its largest reasoning counterpart and runs up to 5× faster than competing models. Two edge‑focused variants—Bolt 74B‑A9B 24GB (native 2‑bit quantized) and Bolt E4B 4GB—extend the family to local RTX/DGX Spark, GeForce RTX, and Jetson platforms, covering deployments from data‑center servers to desktop agents. The models are open for post‑training on proprietary data, enabling organizations to tailor high‑throughput, always‑on agents without sacrificing capability or deployability. Engineers can now select a Nemotron model that matches their compute envelope while retaining state‑of‑the‑art performance.

⚡ Key Takeaways

  • Nemotron 3.5 Super VL: 120 B parameters, achieves up to 5× faster throughput than the next best open model in its class.
  • Bolt 74B‑A9B 24GB uses native 2‑bit quantization to fit within 24 GB memory on NVIDIA RTX/DGX Spark and ranks #1 on PinchBench and #2 on BrowseComp.
  • Bolt E4B 4GB is optimized for GeForce RTX, Jetson, and RTX/DGX Spark, targeting fast conversational agents and embedded interfaces.
  • Availability: Super VL and Bolt 74B‑A9B are available now; Bolt E4B ships on Oct 23.
  • Real‑world post‑training: Scale AI customized Super VL for Skydance’s vendor‑credit finance agent and EY’s M&A due‑diligence agent; AfterQuery adapted it for Hebbia’s credit‑agreement workflows.

🔧 Tools & Libraries

NVIDIA RTXDGX SparkJetsonPinchBenchBrowseComp
💡 Why It Matters

Engineers can deploy open, high‑performance agentic models at any scale—from enterprise servers to edge devices—while retaining full control over data and customization, dramatically reducing latency and cloud‑dependency for always‑on AI services.

✅ Practical Steps

  1. Deploy Bolt 74B‑A9B 24GB on NVIDIA RTX/DGX Spark using its native 2‑bit quantization to stay within a 24 GB memory budget.
  2. Post‑train Nemotron 3.5 Super VL on your proprietary datasets to boost accuracy for specialized agentic workflows.
  3. Run Bolt E4B 4GB on GeForce RTX or Jetson devices to power local voice assistants and interactive characters.

Want the full story? Read the original article.

Read on NVIDIA Blog ↗

More like this

Censored vs. Truncated Regression in Python, Explained

Towards Data Science•#llm

New in AWS Quick: More capable flows, an updated web experience, and agents that work together

AWS ML Blog•#agents

How Google Duplex evolved to also call businesses for you in Gemini

Google Research•#agents

Q&A: Enabling more human-like interactions and understanding in multimodal foundation models

MIT News AI•#llm

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING