Banma Intelligence Launches AutoOmni 2.0 On-Device LLM, Shifting AI Competition to the Edge

At the 2026 Yunqi Conference, Banma Intelligence's end-side large model series AutoOmni welcomed its new flagship model, AutoOmni 2.0-23B-A3B. A car parked outside the venue is the first mass-production vehicle equipped with the Qualcomm Snapdragon 8397 platform and Banma Intelligence's AutoOmni—by Banma Intelligence's definition, its smart cockpit model runs on automotive-grade chips, not in the cloud.

This launch has a policy backdrop. The "AI + Software" Special Action Implementation Plan issued by the Ministry of Industry and Information Technology states that it supports joint innovation between software companies and terminal manufacturers such as smartphones, smart homes, and smart connected vehicles, accelerating the integrated deployment and application of lightweight models and terminal software.

Another set of industry-side judgments comes from Kang Kai, technical director of China Automotive Information Technology (Tianjin) Co., Ltd. At the TEDA Forum, he released the "China Embodied Intelligence Industry Development Report 2026," proposing three judgments, including that embodied intelligence has passed the technology validation stage and is currently in the value validation stage. Banma Intelligence CTO Si Luo, meanwhile, calls smart cars the largest pilot testing ground and large-scale stress-testing platform for embodied intelligence.

The end-side narrative holds only if the hardware is already in place. According to data disclosed by MIIT, in the first half of 2026, the penetration rate of new passenger cars with L2-level combined driving assistance functions reached 70.5%, and the installation rate of models with Navigate on Autopilot (NOA) capability exceeded 30%.

But the payment side is a different picture. According to multiple industry survey data, the overall penetration rate of paid subscriptions and one-time purchases for advanced intelligent driving in China is less than 10%, the paid activation rate for self-developed intelligent driving among most independent brands is less than 30%, and the payment rate for volume-oriented family models below the 200,000 yuan level is even lower than 2%. This data does not specify the specific institutions or samples, and can only be regarded as an industry survey account relayed by the source.

There is a mismatch between recognition and willingness to pay. McKinsey's "2026 China Auto Consumer Insights" shows that about 80% of respondents regard urban NOA as an important option when buying a car.

Banma Intelligence also provided its own business data: according to figures it supplied, since launching a comprehensive AI transformation in mid-2024, its service coverage rate for mainstream automakers has reached 68%, and its market share in end-side intelligence and end-model applications is 63%. These two figures were provided by the company itself and lack third-party statistical corroboration.