iFlytek Unveils On-Device Multimodal Large Model and Proprietary Acoustic System, Driving Intelligent Cockpits Toward Proactive Services

At the Beijing Auto Show, iFLYTEK unveiled its next-generation multimodal intelligent cockpit solution, featuring the SparkAuto-EMM large model deployed on-device, the iFLYSOUND gallium nitride (GaN) master-grade acoustic system, and a multilingual interaction solution tailored for overseas markets. This integrated solution aims to shift from "reactive responses" to "proactive services," establishing a complete closed-loop system encompassing perception, understanding, decision-making, execution, and proactive service delivery. The newly launched on-device multimodal large model supports scalable deployment ranging from 0.5B to 7B parameters, adapting flexibly to various computing platforms. It operates reliably even without network connectivity and ensures user data never leaves the vehicle. Capable of interpreting ambiguous semantics and complex intents—such as automatically deciding whether to open a window or activate the air conditioning upon hearing “It feels stuffy”—the model also enables multi-step task planning. For example, in response to “Refuel first, then head to the airport, and find a Sichuan restaurant along the way,” it can decompose the request, plan the optimal route, and orchestrate relevant services accordingly. Concurrently, iFLYTEK has built a three-tier intelligent agent ecosystem: proprietary high-frequency scenario agents, deeply integrated partner-optimized agents (e.g., Meituan) fine-tuned for in-vehicle use, and extensible solutions for long-tail scenarios. All agents are specifically customized for automotive contexts, prioritizing high first-turn accuracy, minimal interaction steps, and reliable performance under weak or offline network conditions—all while ensuring driving safety. Addressing China’s growing automotive export demands, iFLYTEK’s overseas Spark Assistant supports 32 languages across 60 countries and regions. Leveraging native-speaker-curated corpora and 52 high-quality TTS voices, it integrates local digital ecosystems—including navigation, music, and news—to deliver culturally appropriate interaction experiences. Already, eight of China’s top ten automakers exporting vehicles have partnered with iFLYTEK, with its solutions deployed across more than 100 mass-produced models. In acoustics, iFLYSOUND employs a GaN-based audio amplifier architecture, delivering up to 300W peak output per channel. Combined with ruby film capacitors and low-density finned heat dissipation technology, the system achieves a 20% improvement in sound quality and a 30% reduction in weight—all while maintaining cost efficiency. Certified by QQ Music for its “Premium Sound Quality,” iFLYSOUND is already featured in models such as the WEY V9X, Dongfeng Nissan NX8, Zeekr 007 GT, Chery Exeed EX7, Fulwin T9L, and Hyper S600, with cumulative installations exceeding 1.2 million units. iFLYSOUND also integrates over 30 proprietary audio algorithms, enabling features like zoned sound fields, directional navigation prompts, microphone-free karaoke, and the “On-the-Go DJ” function—delivering a multidimensional experience that is “great-sounding, easy-to-use, and fun.” Notably, the On-the-Go DJ dynamically remixes music in real time based on driving data such as vehicle speed and throttle depth, enhancing the interplay between driving dynamics and audio entertainment. Altogether, this comprehensive solution marks a new phase in intelligent cockpit competition—one defined not by isolated feature stacking, but by system-level intelligent agent capabilities grounded in deep understanding of real-world driving scenarios, automotive-grade engineering implementation, and global-local adaptability.