NVIDIA ships Nemotron 3.5 Bolt 74B for local agents as RTX Spark laptops prep for Friday launch
NVIDIA's open 74B-parameter, 9B-active MoE model, packed into 24GB with 2-bit quantization, is available Thursday for RTX Spark. Agent apps including GitHub Copilot, Hermes and OpenClaw get Spark-ready updates ahead of the Oct. 16 laptop launch.
Photo: Dicklyon / Wikimedia Commons, CC BY-SA 4.0. NVIDIA logo via Wikimedia Commons (Apache License 2.0).
NVIDIA said Nemotron 3.5 Bolt 74B-A9B 24GB becomes available on Thursday, October 15, as a large language model purpose-built for running local agents on RTX Spark, the company's new laptop platform due in stores on Friday, October 16. The announcement is part of an October 13 NVIDIA blog post on the software ecosystem preparing for launch.
NVIDIA describes Bolt 74B as a 74-billion-parameter mixture-of-experts model with 9 billion active parameters, packed into 24GB using 2-bit quantization with quantization-aware training. It says the model aims to balance speed and intelligence for local agents, and that like other Nemotron releases it is fully open source, including training data, weights and recipes.
Users will be able to run the model at launch through llama.cpp, vLLM, Ollama, Unsloth and LM Studio, and directly through agents such as Hermes and OpenClaw, NVIDIA said. Updates to GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy are meant to streamline setup on RTX Spark, improve security with OpenShell and Microsoft Execution Containers, and support hybrid local and cloud routing.
RTX Spark pairs up to 128GB of unified memory with 1 petaflop of AI compute, according to NVIDIA. The same post said NVIDIA PAIR is moving out of beta with llama.cpp support and direct Hugging Face model downloads, and that Llama Windows, a tray app for llama.cpp, is also launching with day-one RTX Spark support.