Intel has launched OpenVINO 2026.0, introducing significant improvements to its open-source AI toolkit. This release expands large language model (LLM) support, notably for GPT-OSS-20B, MiniCPM-V-4_5-8B, and MiniCPM-o-2.6, enhancing functionalities for Intel’s CPU, NPU, and GPU products. Smaller models like Qwen2.5-1B-Instruct and Qwen-2.5-coder-0.5B are now supported on NPUs, while OpenVINO GenAI introduces word-level timestamps for improved transcription and subtitling, positioning it favorably against OpenAI’s solutions. The toolkit also features int4 data-aware weight compression, enhancing the efficiency of MoE LLMs with lower memory and bandwidth needs. Enhanced NPU support includes compiler integration for seamless operation without OEM driver dependencies. Intel’s goal is to streamline integration processes, accelerating time-to-value for AI applications. For further details and downloads, check OpenVINO’s GitHub page. Expect new OpenVINO benchmarks and GenAI performance metrics soon.
Source link
Share
Read more