Overview
In today's digital era, artificial intelligence (AI) is advancing at an unprecedented pace. With its deep investment and pioneering role in AI PCs, Intel® is committed to driving innovation in client AI models. Today, we are excited to announce that in the latest launch of the MiniCPM4 LLM model series, Intel and ModelBest have collaborated deeply during the model development phase to jointly bring new model innovations and AI PC performance experiences to the industry. We believe that through the joint efforts of Intel and ModelBest, the MiniCPM4 series models can be widely adopted for many industry use cases.
This model series boasts innovations in multiple dimensions including architecture, algorithms, and system layers. With a customized speculative decoding configuration through hardware-aware draft model optimization, combined with Intel's acceleration suite and KV Cache memory enhancement technology, the full potential of the hardware is unleashed, achieving an efficiency improvement of up to 2.2X.
This lays a solid foundation for the widespread application and deployment of the MiniCPM4 series model on the Intel platform.
MiniCPM4 is a next-generation large language model designed for efficient deployment, with a continued focus on scalability, lightweight architecture, and strong multilingual and instruction-following performance. The series includes two parameter scales: 0.5B and 8B. Intel's CPU, GPU, and NPU architectures have been fully adapted, ensuring that Intel® Core™ Ultra Processors (Series 2) can leverage the OpenVINO™ toolkit to provide optimized and outstanding performance for the MiniCPM4 series models. Day 0 enablement is also supported on NPU, providing more diverse and targeted platform support for different model parameter sizes and application scenarios.
Figure 1. Throughput of MiniCPM4 models on Intel® Core™ Ultra Processors with built-in GPU
Intel® has made great strides towards enabling long context window use cases on client platforms by leveraging block sparse attention mechanisms, customized operator fusion, and hardware-driven algorithmic optimization.
We successfully supported a 128K long-context window on the latest Intel® Arc™ Pro B60 Graphics while ensuring output quality as shown in Demo 1. This establishes a strong foundation for unlocking more client AI applications. Moving forward, Intel will continue to maintain deep collaboration and joint research and development with ModelBest to further enhance the performance of long-context window applications.
Demo 1. Unlocking 128K long context reasoning capability on Intel® Arc™ Pro B60 Graphics
Summary
Intel AI PCs support MiniCPM4 models at launch and allows developers to further explore MiniCPM4’s potential through use cases such as on-device chatbots, document summarization and Q&A, lightweight code assistance, and multimodal interaction. Intel will continue to maintain close cooperation with a wide range of model manufacturers to drive the continuous development of AI technology and contribute to building a smarter future.
Resources
- Getting started with MiniCPM4 and OpenVINO™
- LLM-powered Chatbot using OpenVINO™ notebook
- OpenVINO™ GenAI Library
- OpenVINO™ Zone in Moda Community
- OpenVINO™ Model Hub for AI Inference Benchmarks
Product and Performance Information
Intel® Core™ Ultra 7 258V Configuration: OEM: Lenovo*, Model: Yoga Air 15s ILL9, CPU: Intel Core Ultra 7-258V, Memory: 32GB LPDDR5-8533MHz, Storage: WD PC SN740 1TB, OS: Windows 11, OS Version: 24H2 (26100.4061), Graphics: Intel Arc 140V GPU, Graphics Driver Version: 32.0.101.6790, Resolution: 2880 x 1800 200% DPI, NPU Driver:32.0.100.4023, Software Version: OpenVINO 2025.2.0-dev20250520, OpenVINO GenAI 2025.2.0.0-dev20250520. Tested by Intel on May 29th, 2025.
Intel® Core™ Ultra 9 285H Configuration: OEM: Lenovo*, Model: Ideapad Pro 5 16IAH10, CPU: Intel Core Ultra 9-285H, Memory: 32GB LPDDR5-8533MHz, Storage: Kioxia KBG60ZNT1T02 1TB, OS: Windows 11, OS Version: 24H2 (26100.4061), Graphics: Intel Arc 140T GPU, Graphics Driver Version: 32.0.101.6790, Resolution: 2880 x 1800 200% DPI, NPU Driver:32.0.100.4023, Software Version: OpenVINO 2025.2.0-dev20250520, OpenVINO GenAI 2025.2.0.0-dev20250520. Tested by Intel on May 29th, 2025.
Notices & Disclaimers
Performance varies by use, configuration and other factors. Learn more on the Performance Index site. Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See backup for configuration details. No product or component can be absolutely secure. Your costs and results may vary. Intel technologies may require enabled hardware, software or service activation.
Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.
AI Disclaimer
AI features may require software purchase, subscription or enablement by a software or platform provider, or may have specific configuration or compatibility requirements. Details available at http://www.intel.com/AIPC. Results may vary.