StepFun and Geely Automobile Group jointly announced that the two Step-series multimodal large models developed through their collaboration will be open-sourced to global developers. Among them are Step-Video-T2V, currently the open-source video generation model with the largest number of parameters and the best performance in the world, and Step-Audio, the industry's first product-grade open-source voice interaction large model. The open-sourcing of the two Step-series multimodal large models brings the industry's strongest multimodal large model capabilities to the open-source world, will significantly promote the sharing and innovation of large model technologies, advance the inclusive development of artificial intelligence, and contribute Geely's solutions and China's strength to the development of the global digital economy.
StepFun is a strategic partner in Geely Automobile Group's technology ecosystem and has a world-class algorithm team. Starting in 2023, Geely began in-depth cooperation with StepFun in areas such as pre-trained large models, vertical-domain large models, application scenario design, and engineering development. Among these, Geely led businesses including scenario design, model evaluation, and engineering development, while StepFun led businesses such as model pre-training. The two sides complemented each other's strengths, significantly enhancing the performance of the two Step-series multimodal large models.
The Step-Video-T2V large model has 30 billion parameters and can directly generate high-quality videos of 204 frames at 540P resolution. It is currently the open-source video generation large model with the largest number of parameters and the best performance in the world. The Step-Audio large model can generate expressions of emotion, dialect, language, singing, and personalized style according to different scenario requirements, and can engage in natural, high-quality dialogue with users. It is the industry's first product-grade open-source voice interaction model.
As a global leader among Chinese automakers, Geely is committed to becoming a leader in and popularizer of AI technology for intelligent vehicles. As early as 2021, Geely built an end-to-end self-developed system and ecosystem alliance around chips, software operating systems, data, and satellite networks, constructing a comprehensive "Intelligent Geely Technology Ecosystem Network" that continuously drives the evolution of user experience in intelligent driving and intelligent cockpits. In 2022, Geely completed the global automotive industry's first Xingrui Intelligent Computing Center integrating "cloud, data, and intelligence." In 2023, Geely released the global automotive industry's first full-stack self-developed full-scenario AI large model, the Xingrui AI Large Model, and received the "4+ level," the industry's highest rating, from the China Academy of Information and Communications Technology. In 2025, Geely released the industry's first "Intelligent Vehicle All-Domain AI" technology system.
With the comprehensive empowerment of the world's strongest open-source Step-series video generation large model, voice interaction large model, and the automotive industry's full-scenario large model, Geely will bring users a higher level of autonomous driving and safer advanced intelligent driving experience across all scenarios such as driving safety, emergency hazard avoidance, and smart parking. At the same time, it will create a more intelligent emotional cockpit for vehicles, redefine the human-vehicle interaction experience, and promote the widespread application of AI technology in the intelligent vehicle field.
The AI Drive large model within Geely's Xingrui AI Large Model has strong scenario generation and planning capabilities and can construct extreme driving scenarios rarely encountered in real driving based on different weather and lighting conditions such as rain and snow, road structures, congested traffic conditions, multi-story garages, and uncommon traffic targets. Optimizing Step-Video-T2V based on real driving video data can turn it into an autonomous driving data synthesizer. The Xingrui AI Large Model is exploring how to convert these rich synthetic scenarios into "indistinguishable-from-real" training data, equivalent to creating a rich "virtual driving test ground" for autonomous driving systems, thereby significantly improving the vehicle's ability to respond to various road conditions during intelligent driving and ensuring driving safety.
Empowered by the super-strong interaction capabilities of the Step-Audio large model, Geely's Xingrui AI Large Model can greatly improve the response speed to voice commands, as well as emotional understanding and expression during interaction. When a user says, "Navigate, I'm going to be late," the system will not only automatically plan a faster route, but also proactively ask whether to play light music to ease the mood. At the same time, the system can not only recognize rich dialects such as Cantonese and Sichuanese, but also support personalized voice customization. Users can choose from the voice timbres provided by the system, record their child's laughter to create a专属 prompt tone, and even have the in-car AI imitate the sound of a mountain stream to create an in-car meditation space, offering warm companionship on every trip.
In addition, Geely is also the first automaker to complete deep integration with DeepSeek. By working with the DeepSeek-R1 model, Geely will conduct distillation training on the Xingrui Vehicle Control FunctionCall large model, the automotive proactive interaction edge-side large model, and others. By then, Geely's intelligent vehicle AI will not only be able to accurately understand users' vague intentions and thereby accurately invoke about 2,000 vehicle interfaces, but also proactively analyze users' potential needs based on in-vehicle and out-of-vehicle scenarios, and proactively provide users with services such as vehicle control, proactive dialogue, and after-sales service. The intelligent interaction experience will be greatly improved, accelerating the popularization of AI in the intelligent vehicle field.
At present, a series of AI applications built on the "Intelligent Vehicle All-Domain AI" technology system, including AI Galaxy Elf, Flyme Auto intelligent cockpit, Wujie Space, AI Cloud Power, AI Picture Books, and AI Sentry, have already been equipped on models such as Galaxy E8, Galaxy E5, and Galaxy Starship 7 EM-i. In early March this year, Geely will also release a major AI intelligence strategy to accelerate the gradual rollout of the latest AI technologies such as AI digital chassis and advanced intelligent driving, bringing users disruptive intelligent experiences.
