Nemotron 3.5 Bolt Hermes: Run NVIDIA’s New Local Model in Hermes Agent
Nemotron 3.5 Bolt Hermes is the pairing of NVIDIA’s newest local model with Hermes Agent. NVIDIA released Nemotron 3.5 Bolt on 15 October 2026, and Hermes Agent v0.22.3 added it to Hermes’ local runtime on 21 October. That means you can run a capable agent model on your own NVIDIA GPU, free and private. In the video below I set up its big brother, Nemotron 3.5 Super VL, as a free model for Hermes. Here’s what Bolt is, what hardware you need and how to run it with Hermes.
Key takeaways
Nemotron 3.5 Bolt: NVIDIA’s 74B-A9B model squeezed into 24GB.
Added to Hermes Agent’s local runtime in v0.22.3.
Text and image input, tool calling, up to 256K context (per Ollama).
Nemotron 3.5 Bolt is part of NVIDIA’s open Nemotron family, announced alongside Nemotron 3.5 Super VL. NVIDIA says the 74B-A9B version uses native 2-bit quantization to fit in 24GB of memory. It’s a mixture-of-experts model with 74 billion parameters in total and about 9 billion active per token, so it runs far faster than its size suggests.
Nemotron 3.5 Bolt at a glance
Made by
NVIDIA
Released
15 October 2026
Versions
74B-A9B (24GB) now; a smaller E4B (4GB) version due 23 October
Input
Text and images, with tool calling and optional thinking (per Ollama)
Context
Up to 256K tokens (per Ollama)
Licence
OpenMDW License Agreement 1.1
Runs with
llama.cpp, Ollama, LM Studio, Unsloth, vLLM; weights on Hugging Face
NVIDIA’s claims
#1 open model in its class on PinchBench, #2 among open models on BrowseComp
Nemotron 3.5 Bolt in Hermes Agent
Hermes Agent v0.22 brought one-click local models, and the v0.22.3 release added Nemotron 3.5 Bolt (plus one-slot DFlash drafts, a speed-up feature) to Hermes’ local runtime. The release before that, v0.22.2, made Nemotron Super 3.5 a default and put a curated model list at the top of the model picker.
For you, that means picking Bolt from the model picker instead of wiring it up by hand. It runs on your own machine, so there are no API bills and your data stays with you.
Want private AI agents running on your own hardware? Inside the AI Profit Boardroom you get the Agent OS — Hermes, OpenClaw, OpenCode, Claude and Codex in one place — plus step-by-step Hermes training and four live coaching calls a week.
NVIDIA’s own numbers are strong for a 24GB model: first among open models in its class on PinchBench and second on BrowseComp. Independent testers are more mixed. MindStudio found Bolt strong at vision and reading scientific diagrams, but weaker at agentic coding and translation.
So I’d use Bolt in Hermes for reading documents, screenshots and charts, research and everyday admin. For heavy coding, I’d pair Hermes with a stronger coding model or hand that work to Claude Code or Codex.
Reading screenshots and documents — invoices, charts, dashboards and PDFs.
Private research — summarise files you don’t want to send to a cloud model.
Everyday admin — triaging messages, drafting replies and planning tasks.
Cron jobs that run often — local models cost nothing per run, so frequent jobs get cheap.
Offline work — keep working when the internet or an API is down.
Match the model to the job: Bolt for private, frequent and visual tasks, and a top cloud model for complex coding. Hermes makes switching models a one-line change.
My verdict on Nemotron 3.5 Bolt Hermes
If you have an NVIDIA GPU with 24GB or more, this is one of the easiest ways yet to run a private, free agent model, now that it’s built into the Hermes local runtime. Use it for vision and everyday tasks, and keep a stronger model for coding. If that fits your setup, Nemotron 3.5 Bolt Hermes is well worth a try.
About Julian Goldie
I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.
FAQ: nemotron 3.5 bolt hermes
What is Nemotron 3.5 Bolt?
NVIDIA’s compact open model released on 15 October 2026: 74B total parameters with about 9B active, quantized to fit in 24GB.
Does Hermes Agent support Nemotron 3.5 Bolt?
Yes — Hermes Agent v0.22.3 added Nemotron 3.5 Bolt to its local runtime, and you can also run it through Ollama.
Can I run Nemotron 3.5 Bolt on a Mac?
Not currently — Ollama says Apple Silicon/Metal isn’t supported. Use an NVIDIA GPU or a different model.
Is Nemotron 3.5 Bolt free?
Yes — the weights are free under NVIDIA’s OpenMDW licence, and running it locally has no API costs.
Is Nemotron 3.5 Bolt good at coding?
NVIDIA reports strong agent benchmarks, but independent tests found it weaker at agentic coding than at vision tasks.
Ready to put Hermes to work in your business? Join the AI Profit Boardroom for the Agent OS, setup training and daily updates.
