At WAIC 2025 in Shanghai, Geely and StepFun unveiled the automotive industry’s first human-like AI agent embedded in the Geely Galaxy M9, alongside a new AI-powered smart cockpit OS and foundational large models, marking a significant leap in embodied intelligence and human-machine interaction in vehicles.

Recently, at the 2025 World Artificial Intelligence Conference (WAIC 2025) in Shanghai, Geely Auto Group, teaming up strategically with StepFun, put on quite an impressive show that really highlighted how fast the world of embodied intelligence is developing in the automotive industry. They showcased a pretty cool lineup of new vehicles, like the Zeekr 9X, LYNK & CO 10 EM-P, Geely Galaxy A7, and especially the Geely Galaxy M9. That last one drew lots of buzz because it introduced the industry’s very first human-like AI agent in a car, which is powered by StepFun’s comprehensive AI voice large model. Honestly, this breakthrough—well, at least in my view—seems like a giant leap forward in how humans and machines interact, with a voice system that sounds surprisingly natural, can pick up on emotional cues, and even change its tone to match what the user’s feeling or trying to communicate.

The demo was just part of it. StepFun also launched its newest foundational AI large model, called Step 3, along with the release of the next-generation smart cockpit operating system, dubbed Agent OS (Preview Version). This OS was developed together with Qianli Technology and Geely Auto Group. Designed especially for AI Agent-type applications, Agent OS offers hyper-natural multimodal interaction, and integrates cloud and edge memory, all woven together with human-machine co-piloting features based on integrated mapping tech. They talk about the operating system creating what they call a ‘third living space’ inside the vehicle—basically, aiming to turn the cockpit into a more natural, human-like environment that feels emotionally engaging and makes the driving experience more intuitive.

Now, Geely’s push into embodied intelligence isn’t just a recent thing; it’s rooted in their early entry into AI, supported by a sizable investment into what they call a “Full-domain AI” system that they officially rolled out at the start of 2025. This system isn’t limited to one part of the vehicle; it spans across its architecture, powertrain, chassis, and cockpit. Even more, it extends into the entire value chain—covering everything from R&D and manufacturing, to after-sales. It’s a bold move that—well, at least in my opinion—sets Geely apart as the only automaker globally to really incorporate AI in a complete, full-spectrum way. This approach underlines their goal to lead in intelligent vehicles. And honestly, their ambitions go beyond just building cars—they aim to evolve into a major robotics company by transforming their vehicles into physical nodes that connect people directly to the expansive AI universe.

Supporting all this are three main AI engines: computing power, algorithms, and data. Along with their partners like StepFun, Qianli Technology, DreamSmart, and Geespace, Geely established the Xingrui Intelligent Data Center 2.0, which offers a combined computing capacity of 23.5 EFLOPS—making it the most powerful among Chinese automakers. When you combine this with the integration of the Xingrui AI large model system with StepFun’s multimodal models, plus a dataset of around 10 trillion tokenized data points and 40 billion pieces of vertical automotive-specific data, it’s clear they’ve built quite a formidable platform for AI innovation and training. This setup is set to support some pretty advanced applications, especially in smart driving: for instance, the upcoming cloud-connected WM+ foundational large model for the Qianli Haohan driver-assistance system, plus an in-vehicle VLA+ end-to-end adversarial safety model expected later this year. These will allow AI-to-train-AI and AI-to-verify-AI processes, meaning safety refinement and verification can be more autonomous and thorough.

Additionally, Geely is a leader in AI integration within manufacturing and vehicle operation. Its existing mass-produced models already incorporate a suite of fully integrated, full-domain AI features. These include the Qianli Haohan system for driving, the Thunderbolt AI Hybrid 2.0 for powertrain management, as well as their AI Digital Chassis and AI Intelligent Architecture platforms. By merging these technologies, the company offers a safer, smarter driving experience—hitting some of the key industry trends, like ecosystem cooperation and operational efficiency, right on the nose.

Their partnership with StepFun isn’t just about product tech, either. They've also gone into open innovation. Together, they've open-sourced two multimodal large models: Step-Video-T2V, for generating high-quality videos, and Step-Audio, which is pretty remarkable because it’s the industry’s first open-source voice interaction model capable of producing multi-style speech synthesis. This move supports developers around the world and helps drive progress in multimodal AI applications—both in automotive tech and outside—and it’s a good example of China’s growing position in the open-source AI scene.

In the bigger picture, China is making hefty efforts to build a self-reliant, domestic AI ecosystem in response to international restrictions. StepFun, along with heavyweights like Huawei and Biren, has joined new alliances designed to coordinate AI chip manufacturing, model development, and infrastructure upgrades. These alliances—shown during WAIC 2025—are a sign of how important China’s local AI industry has become in ensuring independence and innovation in the field.

Li Chuanhai, Vice President and Director of the Automotive Research Institute at Geely, emphasized their rapid strides in AI since they got involved in 2021, and how they’re shifting away from traditional car manufacturing towards comprehensive embodied intelligence. He mentioned the company’s focus on creating an AI architecture that reacts faster and understands more deeply—moving AI from just executing tasks to becoming an intelligent assistant capable of handling complex, multi-layered operations. This vision, I think, aims to turn every Geely vehicle into a kind of intelligent, physical interface connected directly with the AI landscape—a goal that clearly positions them not just as automakers, but as future leaders in robotics and intelligent systems.

All these innovations, along with their strategic collaborations, really put Geely and StepFun closer to the forefront of automotive AI. They’re setting a pretty high bar for integrated, emotionally aware, human-centric vehicle tech, combining deep expertise with a forward-looking vision of smart, emotionally intelligent mobility solutions.

Source: Noah Wire Services