China’s AI Start-ups Rush to Open-Source Models Amid DeepSeek Disruption
- tech360.tv
- Mar 6
- 2 min read
DeepSeek has shaken China’s artificial intelligence (AI) industry with its low-cost models, prompting rival start-ups to launch open-source models and seek fresh funding.
The Hangzhou-based company, a spin-off from a hedge fund owned by Liang Wenfeng, has dominated China’s AI sector, challenging competitors such as Moonshot AI and MiniMax. DeepSeek’s V2 model, released in July 2024, triggered a price war, while its V3 model in December and R1 model in January 2025 raised concerns for many AI firms.
AI analyst Grace Shao, founder of AI Proem, said DeepSeek’s disruption has given China an advantage in AI model capabilities and broadened access to the technology. She noted that while Chinese AI firms had been shifting towards consumer-facing applications, the U.S. market remained focused on enterprise and productivity solutions.
In response, Chinese AI start-ups are working to close the gap with DeepSeek. Beijing-based Zhipu AI, backed by Tsinghua University, announced on Monday that it raised 1 billion yuan (USD 140 million), with support from the Hangzhou municipal government. The company has embraced open-source AI, releasing its latest text-to-image model, CogView-4, which can generate Chinese characters.
Shanghai-based Stepfun, founded in 2023 by former Microsoft Research Asia chief scientist Jiang Daxin, introduced two open-source multimodal models last month: Step-Video-T2V, which generates videos from text, and Step-Audio for voice interactions. The company plans to release an image-to-video model this month. Stepfun’s investors include Tencent Holdings, Qiming Venture Partners, and the Shanghai municipal government-owned Capital Investment Co.
MiniMax, known for its AI apps Talkie and Xingye, joined the open-source movement in January with its large language model MiniMax-Text-01 and multimodal model MiniMax-VL-01. Founder Yan Junjie said he would have chosen open-source from the start if given the chance.
Moonshot AI, the company behind the Kimi chatbot, released its K1.5 multimodal reasoning model in January, coinciding with DeepSeek’s R1 launch. The company introduced open-source architecture and optimiser innovations last month.
Baichuan AI, founded by former Sogou CEO Wang Xiaochuan, has shifted its focus to the medical sector. Earlier this week, the company confirmed it had let go of its financial services team to concentrate on its core medical business.
01.AI, founded by former Google China president Lee Kai-fu, has moved away from training large-scale AI models to focus on industry-specific applications. Earlier this year, it partnered with Alibaba Cloud to establish an industrial AI lab, with several employees transitioning to Alibaba Cloud as part of the collaboration.
DeepSeek’s low-cost AI models have disrupted China’s AI industry, prompting rivals to adopt open-source strategies.
Zhipu AI raised 1 billion yuan (USD 140 million) and released an open-source text-to-image model.
Stepfun introduced open-source video and audio AI models, with plans for an image-to-video model.
Source: SCMP