GTC Spotlights NVIDIA RTX PCs and DGX Sparks Running Latest Open Models and AI Agents Locally

The paradigm of consumer computing has revolved around the concept of a personal device - from PCs to smartphones and tablets. Now, generative AI - particularly OpenClaw - has introduced a new category: agent computers. These devices, like the NVIDIA DGX Spark desktop AI supercomputer or dedicated NVIDIA RTX PCs, are ideal for running personal agents - privately and for free.

NVIDIA GTC, running this week, is showcasing a host of agentic AI announcements including:

New open models for local agents, including NVIDIA Nemotron 3 Nano 4B and Nemotron 3 Super 120B, and optimizations for Qwen 3.5 and Mistral Small 4.

NVIDIA NemoClaw, an open source stack for OpenClaw that optimizes OpenClaw experiences on NVIDIA devices by increasing security and supporting local models.

Easier fine tuning with Unsloth Studio to further improve open model accuracy for agentic workflows.

In-person GTC attendees can swing by the NVIDIA build-a-claw event in the GTC Park, running daily through March 19, from 8 a.m.-5 p.m. NVIDIA experts will help guests customize and deploy a proactive, always-on AI assistant using their device of choice. Whether technical or just curious, participants will name their agent, define its personality and grant it access to the tools it needs - creating a personal assistant reachable from their preferred messaging app.

New Open Models Bring Cloud-Level Quality to Local Agents The next generation of local models - with increasingly large context windows - delivers the intelligence to run agents on PC. Combined with richer user context and powerful local tools, these advances are unlocking new possibilities on AI PCs, especially on DGX Spark, with its 128GB of unified memory that supports models with more than 120 billion parameters.

Nemotron 3 Super, released last week, is a 120 billion parameter open model with 12 billion active parameters, designed to run complex agentic AI systems. Nemotron 3 Super is optimal for powering agents on the DGX Spark or NVIDIA RTX PRO workstations. On PinchBench - a new benchmark for determining how well large language models perform with OpenClaw - Nemotron 3 Super scored 85.6%, making it the top open model in its class.

Mistral Small 4, a 119-billion-parameter open model with 6 billion active parameters - 8 billion including all layers - unifies the capabilities of Mistral's flagship models. Users now have an ultraefficient model optimized for general chat, coding and agentic tasks.

Both of these models run locally on DGX Spark and RTX PRO GPUs.

For GeForce RTX users looking for smaller models, Nemotron 3 Nano 4B is the latest model to join the NVIDIA Nemotron 3 family of open models, providing a compact, capable starting point for building agents and assistants locally on RTX AI PCs. The model is a strong fit for building action-taking conversational personas in games and apps that run on resource-constrained hardware. It's available across any NVIDIA GPU-enabled system and combines state-of-the-art instruction-following and exceptional tool use with minimal VRAM footprint.

In addition, NVIDIA announced optimizations for Alibaba's Qwen 3.5 models, which have demonstrated outstanding accuracy (27B, 9B and 4B) and are suited for running local agents on NVIDIA GPUs. The new models natively support vision, multi-token prediction and a large 262,000-token context window. The dense 27-billion-parameter model excels when paired with an RTX 5090 GPU.

All configurations measured using Q4_K_M quantizations BS = 1, ISL = 1024 and OSL = 128 on NVIDIA RTX 5090 and Mac M3 Ultra desktops. Token generation throughput measured on llama.cpp b7789, using the llama-bench tool. Users can try these models today via Ollama, LM Studio and llama.cpp, with accelerated inference powered by RTX GPUs and DGX Spark. Learn more about the latest on NVIDIA open models.

Faster Creative AI With the Latest RTX-Optimized Models LTX 2.3, Lightricks' state-of-the-art audio-video model, released earlier this month, now has support for NVFP4 and FP8 distilled models, accelerating performance by 2.1x. Learn more about Lightricks' LTX 2.3 model.

In addition, Black Forest Lab's FLUX.2 Klein 9B received an update last week, accelerating image editing by up to 2x. NVIDIA has collaborated with Black Forest Labs to release an FP8 version, optimized for the fastest performance and optimal memory consumption on RTX GPUs.

NVIDIA NemoClaw - NVIDIA Optimizations for OpenClaw AI developers and enthusiasts are buying DGX Spark supercomputers or building dedicated RTX PCs to run autonomous AI agents, such as OpenClaw, that draw context from personal files, apps and workflows and can automate daily tasks. However, as adoption of agentic systems like OpenClaw grows, so do concerns about token costs, as well as security and privacy.

To help address these concerns, NVIDIA this week introduced NemoClaw, an open source stack for OpenClaw that deploys optimizations for OpenClaw on NVIDIA devices. The first features available in NemoClaw are NVIDIA Nemotron open models and the NVIDIA OpenShell runtime. Nemotron local models enable users to run inference locally, which means better privacy and no token costs. OpenShell is the runtime designed for executing claws more safely.

Learn more about NemoClaw. Watch the GTC keynote from NVIDIA founder and CEO Jensen Huang and explore sessions.

Fine-Tuning Made Easy With Unsloth Studio As open models make giant leaps, one way of further improving accuracy is fine-tuning, which allows users to customize a model for their own data and use cases. This technique normally requires in-depth technical expertise, coding knowledge and massive amounts of configuration. Unsloth, a leading open source library for model fine-tuning and alignment, today launched Unsloth Studio, an easy-to-use, web-based user interface that simplifies the fine

LINK:	https://blogs.nvidia.com/blog/rtx-ai-garage-gtc-2026-nemoclaw/...
	See more stories from nvidia

More from Nvidia