quantum chip

Chinese scientists have reported a breakthrough in artificial intelligence research, completing the world’s first fine-tuning of a billion-parameter AI model using a quantum computer. The experiment was conducted on the Origin Wukong system, a domestically developed quantum machine powered by a 72-qubit superconducting chip.

The trial was undertaken at the Anhui Quantum Computing Engineering Research Centre. According to state media reports, the result marks a major step in the integration of quantum computing with large language model (LLM) development, with implications for improving the performance and efficiency of AI systems.

Technical Performance

The fine-tuning experiment used the quantum system to optimise a large-scale AI model, yielding several key performance indicators. In trials involving a dataset focused on mental health chatbot applications, training efficiency improved by 15 per cent. Mathematical accuracy rose from 68 to 82 per cent.

Further gains were observed when the size of the AI model was reduced by 76 per cent. Despite this compression, performance still improved by 8.4 per cent, suggesting that quantum-assisted computing could offer a viable solution to the increasing computational demands associated with large AI systems.

Chinese industry analysts have said the demonstration supports the concept of “lightweighting” in AI—a process whereby models are made more efficient without sacrificing performance. The development could also contribute to easing so-called “computing power anxiety”, a growing concern in AI research due to the intensive hardware requirements of training LLMs.

Capabilities and Industry Uptake

Since its launch in January 2024, Origin Wukong has reportedly processed more than 350,000 tasks, serving sectors including finance, medicine, and engineering. Users from 139 countries have accessed the system through an online platform, according to figures released by Origin Quantum, the company responsible for the machine’s development.

Named after Sun Wukong, the Monkey King from Chinese mythology known for his shape-shifting abilities, the quantum system is intended to symbolise adaptability and multi-domain capability. The Wukong chip is designed to execute multiple tasks simultaneously by harnessing quantum parallelism—a feature that enables substantial acceleration compared to classical processing.

Dou Menghan, Vice President of Origin Quantum Computing Technology Co. and a senior official at the research centre, likened the process to “equipping a classical model with a quantum engine”. He said the synergy between quantum hardware and traditional AI models has the potential to significantly enhance computational efficiency.

Chen Zhaojun, deputy researcher at the Institute of Artificial Intelligence under the Hefei Comprehensive National Science Centre, described the experiment as “a huge step forward” for the application of quantum systems in artificial intelligence.

International Context

The breakthrough places China at the forefront of ongoing international efforts to pair quantum computing with AI development. While Beijing has heavily invested in domestic quantum research and infrastructure, similar initiatives are underway in the United States, Europe, and Canada.

Several Western programmes are pursuing the integration of quantum processors with machine learning frameworks, though to date, few have demonstrated the ability to fine-tune AI models at this scale using operational quantum hardware.

According to the Chinese researchers involved in the Origin Wukong project, this is one of the first known cases in which a quantum computer has executed a full-scale AI fine-tuning workload with measurable results.

AI Fine-Tuning Explained

Fine-tuning refers to the process of adapting a pre-trained AI model for specific use cases or domains. Instead of building a model from scratch, developers apply task-specific datasets to refine the model’s performance. This is common in sectors such as healthcare, finance, and legal services, where accurate domain understanding is critical.

By incorporating relevant language, terminology, and context, fine-tuned models can deliver higher accuracy and more practical outputs in real-world applications. The process also helps reduce development time and associated computational costs.

Quantum computing could further accelerate this trend, offering enhanced speed and energy efficiency when training and adapting models. The Chinese study suggests that quantum processors may be particularly effective in high-dimensional optimisation tasks, a core component of machine learning operations.

Future Outlook

The Chinese team has stated that it intends to expand research in this area, with the aim of applying quantum-enhanced training techniques to larger models and broader industry applications. Analysts expect further developments as competition in the field intensifies.

While current quantum systems remain limited by hardware stability and scalability issues, the success of this experiment is seen as a possible indicator of near-term practical utility in hybrid AI-quantum environments.

As governments and private institutions continue to seek advances in AI capabilities, the integration of quantum processors may become a key feature in next-generation computing platforms. The progress achieved by Origin Wukong could signal a shift in how future AI models are trained, scaled, and deployed.