What NVIDIA Announced
On October 13, 2026, NVIDIA kicked off RTX Spark launch week with a headline aimed squarely at anyone tired of cloud AI bills: Nemotron 3.5 Bolt, a 74-billion-parameter mixture-of-experts model built for local AI agents, releasing October 15. It is a big model that fits in a small box — 74B total parameters with only 9B active per query, compressed to 24GB through 2-bit quantization trained with quantization-aware techniques. And like NVIDIA's other Nemotron releases, it ships fully open source: weights, training data, and training recipes, so developers can fine-tune their own agents on top of it.
The Lineup: Dates and Models
- Nemotron 3.5 Bolt 74B-A9B (24GB): releases October 15, 2026 — the flagship local-agent model
- Nemotron 3.5 Bolt E4B: follows October 23, 2026 — a smaller model for conversational agents and in-game NPCs, demoable in a SkyrimNET Skyrim mod
- RTX Spark laptops: go on sale October 16, 2026
- Inference support at launch: llama.cpp, vLLM, Ollama, Unsloth, and LM Studio, with direct agent access via Hermes and OpenClaw
PAIR, Llama for Windows, and the Local-Agent Stack
The model is only half the story — NVIDIA also shipped the plumbing for a local agent setup. PAIR (Personal AI Router) exits beta on October 13: a free, open-source tool that spreads AI workloads across the PCs on your local network, now with llama.cpp support and Hugging Face downloads. Llama for Windows also launches October 13 — a free open-source app for running open models without touching the command line. Add NVIDIA's new MCP servers for its own apps (NVIDIA App, Broadcast, PAIR, RTX Remix, Omniverse), and agents can control things like Broadcast effects or batch-process files directly.
Creator Tooling: Adobe, ComfyUI, Blender, Broadcast
- Adobe is optimizing Photoshop, Premiere Pro, and Substance 3D for RTX Spark — up to 2x faster AI features, editing, color, and effects (in beta as of October 13)
- ComfyUI adds RTX Video Frame Generation and HDR nodes for AI video workflows
- Blender 5.3 (November 2026) adds DLSS Ray Reconstruction for faster, cleaner renders
- NVIDIA Broadcast 2.3 (October 13): RTX Video Super Resolution, Frame Generation AI effects, Studio Voice profiles, and an offline file processor
| Drop | Date | Why creators care |
|---|---|---|
| Nemotron 3.5 Bolt 74B | Oct 15 | Frontier-class local agent brain, fully open |
| RTX Spark laptops | Oct 16 | Hardware that runs it all locally |
| PAIR out of beta | Oct 13 | Free local workload router for multi-PC setups |
| Adobe RTX Spark optimization | Beta now | Up to 2x faster AI editing in Photoshop/Premiere |
| Broadcast 2.3 | Oct 13 | AI video effects and offline file processing |
What This Means for Prompt and Image Creators
For creators, the real shift is economic and practical: the planning, prompt-refining, and agentic grunt work behind image and video projects can move onto your own desk. Pair a local Nemotron agent (briefing, shot-listing, prompt iteration) with your existing cloud image models for rendering, and the loop gets faster and cheaper. The Adobe and ComfyUI speedups sweeten the same deal — more iterations per hour on the same hardware.
Frequently Asked Questions
What is NVIDIA Nemotron 3.5 Bolt?
It is a 74-billion-parameter mixture-of-experts AI model (74B total, 9B active per query) built for local AI agents, releasing October 15, 2026. It is compressed to 24GB via 2-bit quantization with quantization-aware training, and ships fully open source — weights, training data, and recipes.
What hardware do I need to run Nemotron 3.5 Bolt?
You need a machine that can hold the 24GB quantized model in VRAM — NVIDIA positions it for RTX Spark laptops, which go on sale October 16, 2026. It launches with inference support in llama.cpp, vLLM, Ollama, Unsloth, and LM Studio.
What is NVIDIA PAIR?
PAIR (Personal AI Router) is a free, open-source tool that distributes AI workloads across PCs on your local network. It exits beta on October 13, 2026, adding llama.cpp support and Hugging Face model downloads.
How do the Adobe and ComfyUI announcements affect creators?
Adobe is optimizing Photoshop, Premiere Pro, and Substance 3D for RTX Spark (up to 2x faster AI features and editing, in beta), ComfyUI is adding RTX Video Frame Generation and HDR nodes, and Blender 5.3 (November 2026) gains DLSS Ray Reconstruction — all meaning faster local creative iteration.
Is Nemotron 3.5 Bolt really open source?
Yes — like NVIDIA's other Nemotron models, Bolt ships with open weights, training data, and training recipes, so developers can inspect, fine-tune, and build agents on top of it rather than renting a black-box API.
