RTX Spark laptops get open speech models and a dial for background agents

RTX Spark laptops are on sale from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599. According to Nvidia, each pairs a Blackwell RTX GPU and a Grace CPU on one superchip, with up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance. Early hands-on time with the machines has framed them as the moment the local-AI pitch finally landed on real hardware (a week with RTX Spark), with Windows tuned for them (Windows optimization).

Alongside the laptops, Nvidia introduced Nemotron 3.5 Transcribe, an open-weight speech model for transcription, speaker labels and translation across 40 languages (Transcribe in depth); announced day-zero support for Liquid AI’s LFM3 open models through llama.cpp and vLLM; and highlighted the Nemotron 3.5 Bolt E4B model, which runs multistep agent tasks in as little as 4GB of memory. A 64GB DGX Spark is now available from partners. The update also covers gaming and creative work: Phantom Blade Zero launches with ray tracing and DLSS 4.5, RTX Remix 1.6 brings path tracing to Minecraft Java Edition with agent-assisted modding, and native Photoshop and Premiere support is in beta.

Running capable models on your own machine means sensitive audio, documents and email do not have to leave the device. Transcribe is a good example: Nvidia says one promptable model handles transcription, speaker attribution, translation and questions about the audio, and because the weights are open, developers decide where processing happens. For a clinic or a finance team, that choice matters.

The most telling detail is the GPU resource control in the NVIDIA App. Nvidia is designing for a world where agents run in the background all day while the user works in the foreground, and it is giving people a dial to decide who gets the compute. That quietly acknowledges personal agents are becoming a standing workload rather than a demo, the bet Nvidia made when it lined up the app ecosystem before launch (app ecosystem).

A $2,599 entry price keeps this tier in premium territory. Benchmark claims, such as Transcribe ranking first on the Hugging Face Open ASR leaderboard, are Nvidia’s framing and deserve independent testing. The quantized versions that make Transcribe practical on these laptops are coming soon, not shipping today. And Nvidia highlights LFM3 for always-on agents that triage inboxes and manage calendars. Local is better for privacy than cloud, but an agent with standing access to your mail still has standing access to your mail; permission and monitoring questions do not vanish because the compute sits on a desk. More in our Nvidia coverage, Microsoft coverage and agents coverage.