Intel rolls out AI Playground 3.0 alpha now available
Intel has published AI Playground v3.0.0a, the first alpha of its updated local AI app, with support listed for upcoming platforms such as Panther Lake. The release also brings a redesigned interface and expands the feature set across chat, image, and video workflows.
AI Playground is positioned as an offline, Windows-based suite that runs on Intel Core Ultra systems with Intel Arc graphics, or on Intel Arc discrete GPUs from Series Alchemist or Battlemage with 8GB or more of VRAM. Intel lists 82GB or more of disk space as a baseline requirement once models are accounted for.
The 3.0 alpha adds three new “modalities” to the app. Voice mode enables speech-to-text, using a voice model that runs on the NPU when the option is enabled in settings, then exposes a microphone button in the prompt UI. Vision support adds image input to chat, letting users paste or load an image for analysis. Intel also adds agentic multi-modal tools via MCP tooling, intended to let a single chat session mix text chat, vision chat, and image generation.
Source: Intel
The UI changes are also central to 3.0. AI Playground now uses a single unified prompt window across features instead of separate tabs, adds a light theme option, and introduces Prompt Modes plus expanded presets that apply across Chat, Image Gen, Image Edit, and Video Gen. Intel also calls out a “Pro 2 Image” option that uses the Z-Image-Turbo model for higher-quality image generation.
Under the hood, Intel’s guide describes an installer that sets up the runtime and a backend installation manager that can pull components such as OpenVINO, PyTorch, ComfyUI, and Llama.cpp. Intel notes that this backend setup step requires an internet connection, even though the app’s core usage is framed as offline once models and components are installed.
This is easily one of the best AI tools for discrete GPUs and neither NVIDIA nor AMD has anything this advanced. On the other hand, Intel has not released any open weight models like NVIDIA, so their work relies on 3rd-party offerings.
Source: AI Playground (Github)
