Tech firms up ante on open-source AI models

Chinese companies are doubling down on open-source artificial intelligence-powered models as part of a broader push to bring fast-evolving AI technology to more businesses and developers worldwide, and bolster its application in a diverse range of fields.

Experts said the open-source approach will lower the threshold for the development and application of AI, greatly reduce computing power costs, and foster the sharing of AI tech around the world, as well as boost collaboration and innovation.

The recent progress in open-source large language models has showcased China's growing technological prowess and open attitude in the AI domain, given that Chinese AI startup DeepSeek's open-source models have taken the world by surprise, they added.

Tech heavyweight Alibaba Group has stepped up efforts to enable broad access to its AI technology and innovations by releasing large language models from its Qwen family as open-source, and boasting China's largest AI open-source community platform, ModelScope.

The company has made more than 200 generative AI models open-source in recent years. The models have multimodal capacities and can process and generate various types of content, covering text, images, audio and video.

ModelScope, which was launched in November 2022, hosts over 70,000 open-source models, and the user base has expanded from 1 million in April 2023 to 16 million as of June 30, serving 16 million developers from 36 countries around the world.

It supports developers in experiencing, downloading, fine-tuning, training and deploying models. Various types of open-source AI models have been included in the community.

"We aim to simplify and reduce the cost of developing, customizing and deploying AI models for developers and corporations, thereby enabling the creation of revolutionary AI applications that have a positive impact on society," said Zhou Jingren, chief technology officer at Alibaba Cloud Intelligence, emphasizing they are committed to making AI models more accessible and easier to use.

Baidu Inc has recently open-sourced its multimodal LLM Ernie 4.5 series, consisting of 10 distinct variants. The model family includes mixture-of-experts (MoE) models with 47 billion and 3 billion parameters, the largest model having 424 billion parameters, alongside a 0.3 billion dense model. The Ernie 4.5, launched in March, is Baidu's multimodal foundational model.

The company said the MoE architecture has the advantages of enhanced multimodal understanding and improved performance on text-related tasks. All models are trained with optimal efficiency using the PaddlePaddle deep learning framework, which enables highper-formance inference and streamlined deployment.

Experimental results show that the models achieve state-of-the-art performance across multiple text and multimodal benchmarks, particularly in instruction following, knowledge memorization, visual understanding and multimodal reasoning.

Zhu Keli, founding director of the China Institute of New Economy, said the open-source approach adopted by a string of Chinese AI companies will lower the technical threshold, speed up the popularization of AI tech across various sectors including automobiles, manufacturing, finance and education, and allow more enterprises and developers to participate in AI research and development.

Zhu believes technological innovation is unstoppable, and international cooperation serves as an important way to promote the development of AI tech, adding that China's open and inclusive attitude helps promote the advancement of the global AI industry.

"Open source will allow resource-constrained startups, small businesses and entrepreneurial developers to access cutting-edge AI tech and build their own models more cost-effectively," said Pan Helin, a member of the Expert Committee for Information and Communication Economy, which is part of the Ministry of Industry and Information Technology.

Pan said it will accelerate AI technological advancements and breakthroughs by enabling global developers to create customized industry-specific models, and foster a more competitive and diverse AI ecosystem.

Chinese AI companies have the ability to take the lead in global AI innovation, as they have sought an alternative AI development approach that emphasizes efficiency and open-source collaboration — which is different from their US counterparts — while reshaping the global AI landscape, Pan added.

The market size of the nation's AI sector will reach 1.73 trillion yuan ($241.2 billion) by 2035, accounting for 30.6 percent of the global total, said market research firm CCID Consulting.