Nvidia has released Nemotron Nano v2, a 9-billion-parameter hybrid state-space language model the company says delivers six-times faster inference than similarly sized transformers while improving accuracy. The model’s weights, reasoning checkpoint and a large share of its pre-training corpus are being made available on the Hugging Face platform under an open-source licence that allows commercial use. Nvidia’s internal benchmarks indicate Nemotron Nano v2 surpasses Alibaba’s Qwen3-8B on most standard tests of reasoning and comprehension. The technology underpins the first major upgrade to Project G-Assist, Nvidia’s on-device AI helper for tuning PC gaming rigs. A memory-optimised version of Nemotron Nano v2 cuts VRAM requirements by 40%, enabling G-Assist to run on any GeForce RTX or RTX Pro graphics card with at least 6 GB of memory, including laptops. The updated assistant, plus a new plug-in hub for third-party extensions, will be distributed via the Nvidia app starting 19 August.