We’re excited to announce the new OpenVINO™ 2025.3.0 release, now available for installation directly from the official openSUSE repository. It expands native support for running generative AI and LLM models with acceleration on Intel® CPUs, GPUs, and NPUs!
Highlights of the 2025.3.0 release:
Expanded coverage for GenAI and LLMs:
- New supported models: Phi-4-mini-reasoning, AFM-4.5B, Gemma-3 (1B-it, 4B-it, 12B).
- NPU support for Qwen3-1.7B, 4B, and 8B — with optimized models from the OpenVINO Hugging Face repository.
- Preview support for OpenVINO Execution Provider in Windows ML, leveraging the power of the new Intel® Core™ Ultra line.
Performance and advanced compression:
- Contexts of up to 8K tokens, dynamic prompts, and dynamic LoRA for NPU-based LLMs.
- Dynamic batch size with automatic reshaping and control of multiple concurrent requests.
- New per-channel cache compression technique, in addition to per-token KV-cache, further improving accuracy in GenAI.
Portability and performance anywhere:
- Official support for Intel® Arc™ Pro B50 and B60 GPUs.
- GGUF models in OpenVINO™ Model Server, with support for DeepSeek, Qwen2, Llama 3, etc.
- Data-aware int4 weight compression for ONNX in NNCF: less memory, same accuracy!
With this integration in openSUSE, developers can now take full advantage of the Linux ecosystem to build, test, and deploy AI in a more efficient, portable, and optimized way.
Install via Zypper and take your AI to the next level.
innovatorsforopensuse.org 