At the 2026 Yunqi Conference, XPENG and Banma, a smart car solution provider backed by Alibaba Group and SAIC Motor, announced a joint automotive-grade AI cockpit solution based on an on-device large model, supporting the local deployment and operation of 30B-class large models in vehicles with up to 750 TOPS of effective computing power per chip.
Image Source: Banma
Key Takeaways
The solution combines XPENG's Turing AI chip with Banma' AutoOmni multimodal on-device large model, supporting multi-turn conversations, multimodal interaction and offline knowledge-based Q&A.
XPENG's Turing AI chip delivers 750 TOPS of effective computing power per chip, according to the company, and is designed for automotive, robotics and flying-car applications.
The joint solution adopts an “on-device first, cloud-assisted” architecture, with the vehicle handling core AI functions such as multi-turn dialogue, multimodal interaction and offline inference, while the cloud supports lower-frequency services such as information and weather.
Technology and features
The solution uses XPENG's Turing AI chip as its computing foundation and integrates Banma' AutoOmni multimodal on-device large model. It supports local deployment of 30B-class models and AI cockpit functions including continuous multi-turn dialogue, multimodal perception and interaction, and offline Q&A.
With 750 TOPS of computing power from the Turing AI chip, the 30B-class model can perform real-time inference in the vehicle, according to the source. The solution is designed to support functions including complex task planning, multimodal voice interaction and proactive scenario-based services without relying on a public network.
On-device AI
The joint approach shifts core AI cockpit functions toward on-device processing, while retaining cloud services for lower-frequency functions. According to the source, this architecture is intended to reduce reliance on cloud connectivity and ongoing operating costs.
The source says the on-device approach is also intended to address connectivity and data-transfer issues in overseas markets, with vehicle-side processing keeping cockpit data within the vehicle and reducing reliance on network connections.
Strategic cooperation
The strategic cooperation aims to combine automotive-grade high-computing-power chips with an established in-vehicle OS ecosystem to provide automakers with an integrated “chip + AI OS + large model” cockpit solution. The two companies said they will continue model adaptation and toolchain development following the Yunqi Conference demonstration, with the solution intended for deployment in mid- to high-end vehicle models.
Banma said it has secured in-vehicle deployment cooperation with more than 60% of automakers for its cockpit AI on-device models.
