This quarterly update to the OpenVINO AI toolkit adds support for new large language models (LLMs) for execution on both CPUs and GPUs. Newly-supported models include Phi-4, Mistral-7B-Instruct-v0.3, SD-XL Inpainting 0.1, Stable Diffusion 3.5 Large Turbo, Phi-4-reasoning, Qwen3, and Qwen2.5-VL-3B-Instruct. Additionally, Mistral 7B Instruct v0.3 with OpenVINO can also run atop Intel NPUs.
OpenVINO 2025.2 has also been preparing a GenAI text-to-speech pipeline for the SpeechT5 TTS model, a GGUF reader for seamless integration of Llama.cpp based LLMs, and other enhancements.
OpenVINO 2025.2 also brings new optimizations for Intel Core Ultra Series 2 SoCs as well as Intel Arc B-Series (Battlemage) graphics cards. There is also better Linux support now for Intel Arrow Lake H platforms, the key-value cache compression for CPUs is now enabled by default for INT8, NPU support for FP16-NF4 precision on Intel Lunar Lake SoCs for models up to 8B parameters, and other improvements.
Downloads and more details on the new OpenVINO 2025.2 release via GitHub .