China's AI surge: New models boost open-source, auto tech
A video generated by Step-Video-T2V.
Shanghai-based AI startup StepFun and Geely Auto unveiled two open-source multimodal AI models on Tuesday, propelling China to the forefront of the global AI race following the DeepSeek frenzy.
One of them is the world's largest open-source video generative model and the other an advanced auto voice interaction system.
Clément Delangue, Hugging Face's co-founder and chief executive, reposted the announcement on X, calling StepFun "the next DeepSeek."
The Sora-like Step-Video-T2V is the world's largest and highest-performing open-source video generation model, with 30 billion parameters. It can directly generate 204-frame, 540P resolution videos, ensuring exceptional information density and temporal consistency. It demonstrates strong performance in complex motion, aesthetic human figures, visual imagination, basic text generation, native bilingual (Chinese and English) input, and cinematographic language, according to StepFun.
The Step-Audio large model can generate expressions of emotion, dialects, languages, singing, and personalized styles according to different scene requirements. It can engage in natural, high-quality dialogue with users and is the industry's first product-level open-source voice interaction model, said the company.
Shanghai-based AI startups, including StepFun, MiniMax and Hong Kong-listed SenseTime will showcase upgraded and open-source models at the coming GDC 2025 in Shanghai, a global AI developer conference which opens on Friday.
The advancements, including the DeepSeek wave, underscore China's commitment to AI development. Chinese companies are rapidly closing the gap with global AI powerhouses and building an improved AI ecosystem consisting of open-source and lower AI training and research cost, analysts said.
Elon Musk's artificial intelligence company unveiled on Monday the latest version of its chatbot, Grok 3, which the billionaire hopes will find traction in a highly competitive sector contested by the likes of ChatGPT and China's DeepSeek.
