OpenVINO™ toolkit: An open source AI toolkit that makes it easier to write once, deploy anywhere.

What's New in Version 2025.2

OpenVINO toolkit 2025.2 takes your AI deployments to the next level with new features and performance enhancements. In this release, you’ll see continuous improvements for large language models (LLMs), optimized runtimes for Intel® hardware, and expanded capabilities for efficient AI deployment across edge, cloud, and local environments. Explore the latest updates and unlock new possibilities for your AI projects.

Release Notes

View System Requirements

Latest Features

Easier Model Access and Conversion

Product	Details
New Model Support	New models supported on CPUs & GPUs: 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. Mistral-7B-Instruct-v0.3 is also supported on NPUs.

GenAI and LLM Enhancements

Expanded model support and accelerated inference.

Feature	Details
KV Cache Compression	KV cache compression for CPUs is enabled by default for int8, providing a reduced memory footprint while maintaining accuracy compared to FP16. Additionally, it delivers substantial memory savings for LLMs with int4 support compared to int8.
GenAI Enhancements	Preview: OpenVINO™ GenAI introduces a text-to-speech pipeline for the SpeechT5 TTS model, while the new retrieval augmented generation (RAG) back end offers developers a simplified API that delivers reduced memory usage and improved performance. Preview: OpenVINO GenAI offers a GPT-Generated Unified Format (GGUF) reader for seamless integration of llama.cpp-based LLMs, with Python* and C++ pipelines that load GGUF models, build OpenVINO toolkit graphs, and run GPU inference quickly. Validated for popular models: DeepSeek-R1-Distill-Qwen (1.5B, 7B), Qwen2.5-Instruct (1.5B, 3B, 7B), and Llama-3.2-Instruct (1B, 3B, 8B). Further optimization of LoRA* adapters in OpenVINO GenAI for improved LLM, vision language model (VLM), and text-to-image model performance on built-in GPUs. Developers can use LoRA adapters to quickly customize models for specialized tasks.

Feature

Details

KV Cache Compression

KV cache compression for CPUs is enabled by default for int8, providing a reduced memory footprint while maintaining accuracy compared to FP16. Additionally, it delivers substantial memory savings for LLMs with int4 support compared to int8.

GenAI Enhancements

Preview: OpenVINO™ GenAI introduces a text-to-speech pipeline for the SpeechT5 TTS model, while the new retrieval augmented generation (RAG) back end offers developers a simplified API that delivers reduced memory usage and improved performance.
Preview: OpenVINO GenAI offers a GPT-Generated Unified Format (GGUF) reader for seamless integration of llama.cpp-based LLMs, with Python* and C++ pipelines that load GGUF models, build OpenVINO toolkit graphs, and run GPU inference quickly. Validated for popular models: DeepSeek-R1-Distill-Qwen (1.5B, 7B), Qwen2.5-Instruct (1.5B, 3B, 7B), and Llama-3.2-Instruct (1B, 3B, 8B).
Further optimization of LoRA* adapters in OpenVINO GenAI for improved LLM, vision language model (VLM), and text-to-image model performance on built-in GPUs. Developers can use LoRA adapters to quickly customize models for specialized tasks.

More Portability and Performance

Develop once, deploy anywhere. OpenVINO toolkit enables developers to run AI at the edge, in the cloud, or locally.

Product	Details
Intel® Hardware Support	Optimizations for Intel® Core™ Ultra processor series 2 built-in GPUs and Intel® Arc™ B-series graphics with the Intel® Xe Matrix Extensions (Intel® XMX) systolic platform to enhance the performance of VLM models and hybrid quantized image generation models, as well as improve first-token latency for LLMs through dynamic quantization. Enhanced Linux* support with the latest GPU driver for built-in GPUs on Intel Core Ultra processor series 2 (formerly code named Arrow Lake H). NPU support for FP16-NF4 precision on Intel Core Ultra 200V processor series for models with up to 8 billion parameters is enabled through symmetrical and channel-wise quantization, improving accuracy while maintaining performance efficiency.
Model Server Updates	OpenVINO™ model server now offers a streamlined C++ version for Windows* and enables improved performance for long-context models through prefix caching and a smaller Windows package that eliminates the Python dependency. Support for Hugging Face* models is now included.
NNCF Updates	Support for int4 data-free weights compression for ONNX* (Open Neural Network Exchange) models implemented in the Neural Network Compression Framework (NNCF).

Sign Up for Exclusive News, Tips & Releases

Be among the first to learn about everything new with the Intel® Distribution of OpenVINO™ toolkit. By signing up, you get early access product updates and releases, exclusive invitations to webinars and events, training and tutorial resources, and other breaking news.