Developers and AI enthusiasts can now run capable AI models and agents directly on their own NVIDIA hardware, keeping private files, code, and project data on the device while getting fast responses. Throughout August 2026, NVIDIA released a run of open-weight local AI models, developer tools, and hardware optimizations built for RTX GPUs, DGX Spark, DGX Station, and Jetson. The releases center on agents that read local data, use tools, and work through long, multistep tasks without depending on the cloud.
What did the August open source push cover?
The updates span new open models, applications, and tooling, along with the accelerated computing libraries and educational resources that help users get started. The common thread is local operation: models sized to run on a single consumer GPU or a personal AI system, with optimizations for NVIDIA hardware so agents can stay always on.
How does Perplexity’s Portable Computer run agents locally?
Perplexity’s Portable Computer app lets users run a personal AI agent on NVIDIA DGX Spark for local, private workflows that do not count toward token limits. It connects to apps including Google Drive, Gmail, Slack, and GitHub, and lets users switch between local models for everyday tasks and cloud models for more complex work. Running locally, it offers one-click inference setup and uses a specially post-trained Qwen 3.8 27B model, with DGX Spark providing fast inference and 24/7 always-on operation for long-running autonomous agents. A fine-tuned Nemotron 3.5 Lightning variant for ultra-fast responses, plus Windows and DGX Station support, are in progress, and support for GeForce RTX and RTX PRO GPUs is coming soon.
Qwen3.8-27B brings coding agents to RTX GPUs
Qwen3.8-27B is a 27-billion-parameter open model sized for a single GPU and built for responsive coding workflows with local files, tools, and project context. It reaches 131 tokens per second on a single GeForce RTX 5090 using multi-token prediction, giving fast local inference for agentic coding, tool use, and longer development tasks. Developers keep sensitive code and proprietary data on their own systems while using the model through llama.cpp, Ollama, Unsloth, and LM Studio Bionic. Day-zero support covers NVIDIA RTX PRO GPUs and the GeForce RTX 5090, and the model also runs on DGX Spark, DGX Station, and Jetson, with NVFP4 optimizations coming.
The benchmark was measured on a GeForce RTX 5090 with an Intel Core Ultra 9 285K, 64GB RAM, Windows 11, and NVIDIA driver 610.43, using the Q4_K_M checkpoint on llama.cpp.
New open-weight models accelerated by NVIDIA GPUs
Several open models became available for local use across NVIDIA hardware:
- Cosmos 3 Edge: a 4-billion-parameter open world model for robotics, autonomous vehicles, and vision AI, running on device on DGX Spark and Jetson at a quarter the size of Cosmos 3 Nano.
- MiniMax-H3: a 33-billion-parameter open-weight model that generates video and natively synchronized stereo audio from text, images, video, audio, or a mix, accessible through ComfyUI.
- Poolside AI Laguna S 2.1: a 118-billion-parameter open-weight agentic coding model for hours-long tasks, with an NVFP4 checkpoint that runs on a single DGX Spark at lower compute and memory cost.
- DeepSeek-V4-Flash: a 284-billion-parameter mixture-of-experts model with 13 billion active parameters and a 1 million-token context window, runnable locally on a DGX Station via community-built GGUF versions.
- Thinking Machines Lab Inkling-Small: a 276-billion-parameter open-weight multimodal model with reasoning across text, images, and audio, activating 12 billion parameters per token. Trained on NVIDIA GB300 NVL72, it runs on a single DGX Station or two DGX Spark systems, with an NVFP4 checkpoint on Hugging Face.
- Unsloth Desktop: a fully open source desktop app that both trains and runs AI models locally, adding image and video diffusion, fine-tuning, agent integrations, web research, and code execution.
- Alibaba Wan-Animate-2: a 14-billion-parameter open-weight model that transfers motion and facial expressions from a driving video onto a static character image, with day-zero ComfyUI support and generation up to 26x faster on RTX PRO 6000 Blackwell and 22x faster on RTX 5090 versus Apple M3 Ultra.
Meta’s Muse Glimmer for always-on local agents
Meta released Muse Glimmer, a 30-billion-parameter dense open-weight model with a 120K-plus context window built for coding and local agentic AI. It delivers over 200 tokens per second on an RTX 5090, and its dense architecture with hybrid attention keeps processing and memory demands manageable as agents take on longer tasks. Small enough to run on a PC with a single consumer GPU, it supports custom agents, private data processing across local files and messages, credential handling with inference kept on the device, long multistep tool calls, and resumable long-running workflows. Developers can run it with vLLM or llama.cpp, use DFlash speculative decoding for faster generation, build agents with NemoClaw, and fine-tune locally with NVIDIA NeMo Automodel.
LTX-2.5 for local video generation
LTX-2.5 is an open-world video generation model delivering higher-quality video, stronger prompt adherence, and more consistent characters, scenes, and voices across generations. New multishot support generates sequences that span multiple cuts while keeping continuity, and an upgraded diffusion video decoder improves fidelity and reduces artifacts. Its prompt enhancer uses Gemma4 E2B and a custom Gemma4 12B text encoder for tighter control over output. On an RTX 6000 PRO GPU, LTX-2.5 runs up to 20% faster with 40% memory savings, and creators can use it locally through a ready-to-use ComfyUI workflow for text, image, and video-to-video generation.
Clustering and managing DGX Spark systems
Larger open models such as GLM 5.2 and DeepSeek-V4-Flash can require multiple GPUs or DGX Spark systems working together. NVIDIA Sync, available for Windows and macOS, detects connected systems, provides secure remote access through Tailscale, and launches applications across one or more DGX Spark systems without manual networking. Its Cluster Assistant configures two or more DGX Spark units as a high-speed cluster over NVIDIA ConnectX-7 ports, routing workloads across nodes and monitoring system health. Later in August, DGX Spark also gains a native ARM64 Linux build of Google Chrome installable in one click, plus a new NVIDIA Sync Resource Monitor that shows real-time and historical CPU and GPU usage across a single system or a full cluster.
Nemotron 3.5 Lightning for fast agentic tasks
NVIDIA expanded its Nemotron 3 family with Nemotron 3.5 Lightning, a customizable open 30B mixture-of-experts model for always-on agents. It delivers up to 4x faster token generation and 30% faster time to completion compared with open models in its class. Because it is open-weight, developers can fine-tune it with their own examples to match specific tasks, such as writing in a preferred tone and format or learning a specialty.
FAQ
What is NVIDIA DGX Spark used for in local AI?
It runs open models and AI agents locally with fast inference and 24/7 always-on operation, suited to long-running autonomous agents. Multiple DGX Spark systems can be clustered with NVIDIA Sync and ConnectX-7 ports to run larger models.
How fast is Qwen3.8-27B on a GeForce RTX 5090?
It reaches 131 tokens per second on a single GeForce RTX 5090 using multi-token prediction, giving responsive local inference for agentic coding, tool use, and longer development tasks.
What can Meta’s Muse Glimmer do?
Muse Glimmer is a 30-billion-parameter dense open-weight model with a 120K-plus context window for coding and local agents. It runs at over 200 tokens per second on an RTX 5090 and supports custom agents, private data processing, credential handling on device, and resumable multistep workflows.
Related coverage
- Cisco Antares: Open-Weight AI Models for Vulnerability Localization
- White House Reportedly Moves to Restrict Open-Source AI
- Qwen3.6-27B: Open-Source 27B Model Beats 397B MoE on Coding Benchmarks
Want help with this in Myrtle Beach or Long Island? Talk to us.
NewSunSEO
Grow in Google and AI Search
AI-SEO, local search, paid ads and websites from one team, with offices on Long Island and in Myrtle Beach.
- AI-SEO
- Local AI-SEO
- AI Consulting
- PPC
- Web Design
- Web Development
- Long Island Office(631) 320-1700
- Myrtle Beach Office(843) 492-7131
Request a Free Proposal
Tell us about your business and where you want to grow.
We will reply by email to talk through your goals and what a proposal would include.
We use your details only to reply to you and send what you asked for. See our Privacy Policy.
