As artificial intelligence (AI) technology advances rapidly, large models are emerging as the key drivers of industrial intelligence upgrades. In the energy and chemical sector—a cornerstone of the national economy and people’s livelihoods—CNPC’s self-developed Kunlun Model has led the way by becoming the first to pass industry-wide accreditation. This milestone underscores CNPC’s commitment to building a secure, independent, and localized AI technology foundation while spearheading intelligent transformation across the industry.

On May 28, CNPC officially unveiled the latest iteration of the Kunlun Model, introducing six industry-leading AI capabilities and achieving large-scale deployment across 152 application scenarios spanning the entire industrial chain. This marks a pivotal shift for the Kunlun Model: from general question-and-answer functionality to proactive intelligence, evolving from “passive response” to “active initiative,” from “pilot testing” to “full industrial chain integration,” and from “reliance on external computing power” to “complete autonomy and controllability.” As the first large model platform in China’s energy and chemical sector to achieve systematic, large-scale implementation across the entire industrial chain, it sets a new standard for independent innovation and high-quality AI-driven industrial development in China’s energy landscape.

The Kunlun Model is the first large model in China’s energy and chemical industry to achieve industry accreditation. Built on a “1+4+N” architecture, it establishes a secure, independent, and fully stacked localized AI technology foundation. The model has pioneered dual alignment of accuracy and performance for 21 large models, including DeepseekV4, on domestic Ascend chips. It also introduced the first long-text acceleration architecture for the Ascend environment, effectively resolving challenges such as computing power overload-induced stuttering and inefficiency.

Currently, CNPC’s AI middleware orchestrates 1,754P of intelligent computing power and stores 620TB of high-quality training data for the energy and chemical sector. According to authoritative evaluations by the China Electronics Standardization Institute, the data quality score reaches an impressive 99.8 points, laying a solid foundation for the stable, efficient, and secure operation of AI systems.

The six advanced AI capabilities unveiled by the Kunlun Model encompass autonomous planning and task decomposition, tool scheduling and execution, professional computing engines, predictive warning engines, inversion computing, and explainable analysis engines. These capabilities fully integrate into the entire industrial workflow—from problem identification, analysis, and decision-making to automated execution and result feedback—effectively transforming AI from a passive responder (“you ask, I answer”) into an intelligent assistant capable of “autonomous thinking and proactive initiative.”

The Kunlun Model has deeply integrated into CNPC’s core operations, establishing 152 application scenarios across three key domains: technological innovation, industrial development, and management enhancement. These scenarios span critical links throughout the industrial chain, including oil and gas exploration and development, refining and production, technical services, and capital finance, leading the industry in both scale and depth of implementation.

In technological innovation, the Kunlun Model has revolutionized traditional petrochemical R&D models by pioneering an “AI-driven simulation and prediction, followed by human experimental validation” approach. The intelligent acoustic full-waveform application scenario has advanced from 2D to 3D, reducing processing time from 20 days to just 3 days while cutting overall costs by over 30%. It represents the industry’s first 3D inversion large model application to achieve industrial-grade accuracy. Additionally, the groundbreaking synthetic rubber large model accurately predicts the performance of seven core materials, with key property prediction accuracy reaching up to 95%, significantly reducing repetitive experiments and saving both time and costs.

In industrial development, newly established risk warning and response scenarios for blowouts, leaks, and stuck drill pipes have achieved over 85% accuracy in drilling risk warnings, with over 300 cumulative warnings issued in the first six months of operation. The plunger lift intelligent diagnosis and optimization scenario has been fully implemented in the Changqing Oilfield, enabling intelligent parameter adjustment for over 3,000 wells and reducing workload by 67% compared to traditional manual management.

In terms of management enhancement, the model incorporates the expertise of 172 specialists, including “National Craftsman” Liu Li, whose “digital avatars” provide round-the-clock business guidance. Meanwhile, the “Oil Treasure” intelligent agent possesses 76 general skills and, in collaboration with experts, has developed 31 professional skills to assist employees with schedule management, document organization, content optimization, and other tasks, effectively bringing AI from “dialog boxes” to the “industrial frontlines.”

It is reported that the simultaneously launched international version of the Kunlun Model supports seven languages: Chinese, English, French, Russian, Arabic, Spanish, and Portuguese. It is the first large model in China’s energy and chemical industry capable of delivering multilingual industrial intelligent services.

This version is now open for global use in the energy and chemical sector, serving not only CNPC’s overseas business operations but also providing multilingual AI services for international energy companies and overseas partners across full-scenario production control, intelligent R&D, safety warnings, and operations management. It facilitates the global expansion of China’s independently developed energy AI technologies and empowers central SOEs to “go global.”