Zebra Intelligent releases AutoOmni 2.0 for on-device automotive AI
The AMW Read
The release incrementally advances an automotive model provider's on-device capabilities, but its broader impact remains unproven without independent evaluations or demonstrated customer economics.
Zebra Intelligent releases AutoOmni 2.0 for on-device automotive AI
Zebra Intelligent unveiled AutoOmni 2.0-23B-A3B at the September 23 Apsara Conference, introducing a multimodal mixture-of-experts model with 23 billion total parameters and 3 billion active parameters. The company said routine-task performance approached that of cloud models with ten times as many parameters, while complex-task performance reached 80%-90% of that comparison group. The report provides no independent evaluation of those claims. Demonstrations included wake-word-free interaction, on-device WeChat operations, visual passenger greetings and parking payments.
The release puts automotive AI at the intersection of model efficiency and agent orchestration. Zebra's proposed advantage is continuous local perception: retaining context and responding quickly while assigning more demanding work to cloud models. Its unified device-cloud agent framework addresses which tasks run locally and which summarized memories reach the cloud. That makes persistent context and service integration potential sources of differentiation alongside model capability. The parking demonstration illustrates the ambition: identify a payment code visually and complete a transaction without a parking-operator partnership.
For builders and investors, the concrete test is whether these experiences justify the hardware bill. Conference participants said higher-end cockpit chips and memory could add several thousand yuan in hardware costs. Buyers should compare that expense with repeat usage and reliable task completion under weak connectivity. Zebra's executives emphasized daily engagement, session duration and usage scenarios; those measures would help establish whether continuous interaction creates enough value to support adoption beyond demonstrations.
