Gemma 4:\u00a0Compact\u00a0Models Optimized for NVIDIA GPUs\u00a0<\/p>\n
The\u00a0latest additions to the\u00a0Gemma 4\u00a0family\u00a0of open models\u2014\u00a0spanning E2B, E4B, 26B and 31B variants\u00a0\u2014\u00a0are\u00a0designed for efficient deployment from edge devices to high-performance GPUs.\u00a0\u00a0<\/p>\n
This new generation of compact models\u00a0supports\u00a0a range of tasks,\u00a0including:\u00a0<\/p>\n Reasoning:\u00a0Strong performance\u00a0on complex problem-solving tasks.\u00a0\u00a0 The\u00a0E2B and E4B models are built for ultraefficient, low-latency inference at the edge, running completely offline with near-zero latency across many devices including Jetson Nano modules.\u00a0<\/p>\n The\u00a026B and 31B modelsare designed for high-performance reasoning and developer-centric workflows, making them well suited for agentic AI.\u00a0Optimized\u00a0to deliver\u00a0state-of-the-art, accessible reasoning, these models run efficiently on NVIDIA RTX GPUs\u00a0and\u00a0DGX Spark\u00a0\u2014\u00a0powering development environments, coding\u00a0assistants\u00a0and agent-driven workflows.\u00a0\u00a0<\/p>\n As local\u00a0agentic\u00a0AI\u00a0continues to gain\u00a0momentum,\u00a0applications\u00a0like\u00a0OpenClaw\u00a0are\u00a0enabling always-on AI assistants\u00a0on RTX PCs,\u00a0workstations\u00a0and DGX Spark.\u00a0The\u00a0latest\u00a0Gemma 4\u00a0models\u00a0are\u00a0compatible with\u00a0OpenClaw,\u00a0allowing\u00a0users\u00a0to\u00a0build capable local agents that draw context from personal files,\u00a0applications\u00a0and workflows to automate tasks.\u00a0Learn how to run\u00a0OpenClaw for free on RTX GPUs and DGX Spark<\/a>\u00a0or\u00a0using\u00a0the\u00a0DGX Spark OpenClaw playbook<\/a>.\u00a0<\/p>\n Getting\u00a0Started:\u00a0Gemma 4\u00a0on\u00a0RTX\u00a0GPUs\u00a0and DGX Spark\u00a0<\/p>\n NVIDIA\u00a0has collaborated with\u00a0Ollama\u00a0and\u00a0llama.cpp\u00a0to provide the best local deployment experience for each of the Gemma 4 models. \u00a0\u00a0\u00a0<\/p>\n To\u00a0use\u00a0Gemma 4 locally, users can\u00a0download Ollama\u00a0to run Gemma 4 models\u00a0or\u00a0install\u00a0llama.cpp\u00a0and pair it with the\u00a0Gemma 4 GGUF Hugging\u00a0Face checkpoint.\u00a0Additionally,\u00a0Unsloth\u00a0provides day-one support with\u00a0optimized and quantized models\u00a0for\u00a0efficient local fine-tuning and deployment\u00a0via\u00a0Unsloth\u00a0Studio.\u00a0Start\u00a0running\u00a0and\u00a0fine-tuning\u00a0Gemma 4\u00a0in\u00a0Unsloth\u00a0Studio today.\u00a0<\/p>\n Running open models like the Gemma 4 family on NVIDIA GPUs\u00a0achieves\u00a0optimal\u00a0performance\u00a0because\u00a0NVIDIA Tensor Cores accelerate AI inference workloads\u00a0to deliver\u00a0higher throughput and lower latency for local execution.\u00a0Plus, the CUDA software stack ensures broad compatibility across leading frameworks and tools, enabling new models to run efficiently from day one.\u00a0\u00a0<\/p>\n This combination allows open models like Gemma 4 to scale across a wide range of systems \u2014 from Jetson Orin Nano at the edge to RTX PCs, workstations and DGX Spark \u2014 without requiring extensive optimization.\u00a0<\/p>\n Check out\u00a0the\u00a0NVIDIA technical blog<\/a>\u00a0for more details on how to get started with Gemma 4 on NVIDIA GPUs\u00a0and\u00a0learn more about\u00a0NVIDIA\u2019s\u00a0work\u00a0on\u00a0open\u00a0models<\/a>.\u00a0<\/p>\n #ICYMI: The Latest Updates\u00a0for RTX AI PCs\u00a0<\/p>\n \u2728\u00a0Catch up on\u00a0RTX AI Garage<\/a>\u00a0blogs\u00a0for a host of agentic AI announcements\u00a0from NVIDIA GTC,\u00a0such as\u00a0new open models for local agents.\u00a0These models include\u00a0NVIDIA\u00a0Nemotron\u00a03 Nano 4B and\u00a0Nemotron\u00a03 Super 120B, and optimizations for Qwen 3.5 and Mistral Small 4.\u00a0<\/p>\n \u00a0NVIDIA\u00a0recently\u00a0introduced\u00a0NVIDIA\u00a0NemoClaw,<\/a>\u00a0an\u00a0open\u00a0source\u00a0stack\u00a0that\u00a0optimizes\u00a0OpenClaw\u00a0experiences on NVIDIA devices by increasing security and supporting local models.\u00a0\u00a0<\/p>\n \ud83d\ude80\u00a0Accomplish.ai<\/a>\u00a0announced Accomplish FREE, a no-cost version of its\u00a0open\u00a0source\u00a0desktop AI agent with built-in models. It\u00a0harnesses NVIDIA GPUs to run open\u00a0weight models locally, while a hybrid router dynamically balances workloads between local RTX hardware and the cloud\u00a0\u2014\u00a0enabling fast, private, zero-configuration execution without requiring an\u00a0application programming interface\u00a0key.\u00a0<\/p>\n Plug in to NVIDIA AI PC on\u00a0Facebook<\/a>,\u00a0Instagram<\/a>,\u00a0TikTok<\/a>\u00a0and\u00a0X<\/a>\u00a0\u2014 and stay informed by subscribing to the\u00a0RTX AI PC newsletter<\/a>.\u00a0<\/p>\n Follow NVIDIA Workstation on\u00a0LinkedIn<\/a>\u00a0and\u00a0X<\/a>.\u00a0\u00a0<\/p>\n \t\t![\"\"](\"https:\/\/www.newsbeep.com\/nz\/wp-content\/uploads\/2026\/04\/gemma-4-perf-chart-desktop-light-1.png\")
<\/a>All configurations measured using Q4_K_M quantizations BS = 1, ISL = 4096 and OSL = 128 on NVIDIA GeForce RTX 5090 and Mac M3 Ultra desktops. Token generation throughput measured on llama.cpp b7789, using the llama-bench tool.<\/p>\n
\nCoding:\u00a0Code generation and\u00a0debugging for\u00a0developer workflows.\u00a0\u00a0\u00a0
\nAgents:\u00a0Native support for structured tool use (function calling).\u00a0\u00a0
\nVision, Video and Audio\u00a0Capabilities:\u00a0Enables rich multimodal interactions for object recognition, automated speech recognition,\u00a0and\u00a0document\u00a0or\u00a0video\u00a0intelligence.\u00a0
\nInterleaved Multimodal Input:\u00a0Mix text and images in any order within a single prompt.\u00a0\u00a0
\nMultilingual:\u00a0Out-of-the-box support for 35+ languages, pretrained on 140+ languages.\u00a0<\/p>\n