Banma Intelligent launches AutoOmni 2.0 edge AI model for always-on smart cockpits
At the 2026 Yunqi Conference in Hangzhou, Banma Intelligent (an Alibaba subsidiary) released its AutoOmni 2.0-23B-A3B full-modal on-device large model, claiming task performance comparable to cloud models with 10 times the parameter count. The MoE-architecture model enables always-on, wake-free in-car interaction. The company also unveiled the AutoClaw 2.0 cockpit service and reported its Yuanshen AI now covers 68% of mainstream automakers. Banma positions smart vehicles as a testbed for embodied intelligence.
Editorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Banma Intelligent Unveils AutoOmni2.0, Eyes Expansion from Smart Cars to Embodied AI
At the 2026 Yunqi Conference in Hangzhou, Banma Intelligent (a subsidiary of Alibaba Group) released its new full-modal on-device large model AutoOmni2.0-23B-A3B and the AutoClaw2.0 smart cockpit collaboration service. The company signaled a strategic shift toward embodied AI, using the mass-produced smart car as a 'pilot site' to train AI in understanding users, environments, and tasks before extending these capabilities to robots and other terminals. Banma Chief Product Officer Cai Ming argued that on-device AI's core value is enabling a new interaction paradigm—shifting from 'wake-up' to 'always-on' presence. The company also noted that about 20% of its revenue now comes from post-sale service and content operations. The move aligns with Alibaba Group's broader AI strategy, as CEO Wu Yongming emphasized AI moving from digital to physical worlds. Banma aims to evolve from a smart cockpit provider to a Personal AI platform, connecting cloud and vehicle, multiple user devices, and services.
Read source2026 Yunqi Conference: 23B MoE Edge AI Model Deployed in Vehicles, Reshaping Cockpit Interaction
At the 2026 Yunqi Conference, Banma Intelligence launched AutoOmni2.0-23B-A3B, a new full-modal edge-side large model using a Mixture of Experts (MoE) architecture with 23B total parameters and 3B activated parameters. The model aims to bring cloud-level intelligence to vehicle cockpits, claiming task performance comparable to 10x larger cloud models. Banma executives argued that edge AI's core value is not just latency or privacy, but enabling an 'always-on' interaction paradigm that replaces wake-word-based systems. They reported that Banma's Yuanshen AI covers 68% of mainstream automakers and that 63% of edge-model-equipped vehicles use Banma's solution. The company also unveiled the AutoClaw2.0 cockpit collaboration service and a partnership with YoooClaw to connect in-car and out-of-car user contexts. Banma executives forecast that 90% of essential interactions will be handled on-device, and that revenue from service and content operations could rise from 20% to 50% of total revenue. They positioned smart vehicles as the largest 'middle-testing ground' for embodied intelligence, with potential to scale to other terminals.
Read sourceBanma Intelligence Launches Full-Modal Edge AI Model AutoOmni 2.0, Says Smart Cars Are Key Testbed
During the Yunqi Conference on September 23, Banma Intelligence (Banma Zhineng) launched its new full-modal edge-side large model, AutoOmni 2.0-23B-A3B. According to the company, the model's task processing capability is comparable to a cloud model with 10 times the parameter count. Banma Intelligence co-CEO Hao Fei stated that the core value of edge intelligence is enabling a new interaction paradigm, with key elements including full-vehicle, always-on wake-free operation, context, and memory, allowing for proactive services. The model uses a Mixture of Experts (MoE) architecture. Chief Technology Officer Si Luo announced that AutoOmni has partnered with 10 domestic and international chip companies, achieving 5-6x inference acceleration, over 99% quantization fidelity, and over 50% reduction in memory usage. A live demo showed the model running on the Freelander 8 vehicle, the first production car based on the Qualcomm 8397 platform. Si Luo emphasized that smart cars are the largest pilot testing ground for embodied intelligence, and Banma is evolving from AutoOmni to EdgeOmni, expanding from single to group intelligence.
Read sourceShow 5 older updatesHide older updates
Zebra Smart launches AutoOmni2.0 AI model to make in-car AI understand user's world
At the Yunqi Conference on September 23, Zebra Smart (Banma Intelligent) released its new full-modal end-side large model, AutoOmni2.0-23B-A3B. The company claims its task processing capability is comparable to cloud models with 10 times the parameter count. Zebra Smart Co-CEO Hao Fei stated that the core value of end-side intelligence is enabling a new interaction paradigm, with full-vehicle, always-on wake-free functionality and context/memory as key elements. Chief Product Officer Cai Ming explained that the 'Yuan Shen AI' system aims to understand both the physical world (the 'table world') and the user's personal intentions, preferences, and life scenarios (the 'inner world'). The company also showcased the AutoClaw2.0 smart cabin collaboration service solution, developed with YoooClaw, which connects cloud and vehicle computing, multiple user terminals, and cross-app services. Chief Technology Officer Si Luo demonstrated features like always-on wake-free, active smart welcome, and seamless parking payment. The model uses a MoE architecture and supports 8+ concurrent tasks with 5-6x inference acceleration. Zebra Smart positions smart cars as the largest 'pilot field' for embodied intelligence, and plans to evolve from AutoOmni to EdgeOmni, expanding from single to group intelligence.
Read sourceBanma Intelligent Launches AutoOmni 2.0 End-Side Model for Smart Cockpits
At the Alibaba Cloud Apsara Conference on September 23, Banma Intelligent (Banma Zhineng) released its new full-modal end-side large model, AutoOmni 2.0-23B-A3B, which it claims achieves task processing capabilities close to cloud models with ten times the parameter scale. The company also showcased the AutoClaw 2.0 smart cockpit collaboration service in a real vehicle. Banma Intelligent Co-CEO Hao Fei stated that the core value of end-side intelligence is enabling a new interaction paradigm, including full-vehicle, always-on wake-free interaction and contextual memory for proactive services. Chief Product Officer Cai Ming emphasized that AI must understand both the physical world and the user's personal world to succeed. The company announced that its Yuanshen AI now covers 68% of mainstream automakers. Chief Technology Officer Si Luo demonstrated scenarios such as always-on wake-free interaction and intelligent parking payment, and noted that AutoOmni has partnered with 10 chip companies globally. Banma Intelligent positions smart cars as the largest pilot field for embodied intelligence.
Read sourceZebra Smart launches AutoOmni 2.0 edge AI model for smarter car cockpits
At the Alibaba Cloud Apsara Conference on September 23, Zebra Smart (Banma Intelligent) released AutoOmni 2.0-23B-A3B, a new full-modal edge-side large language model. The company claims the model, using a Mixture of Experts (MoE) architecture, can handle tasks with performance comparable to cloud models ten times its parameter size. Zebra Smart co-CEO Hao Fei stated that the core value of edge intelligence is enabling a new interaction paradigm through always-on, wake-free voice control and contextual memory. Chief Product Officer Cai Ming explained that the company's 'Yuanshen AI' aims to understand both the physical world and the user's personal world, evolving from cockpit AI to Personal AI. The system connects cloud and vehicle computing, multiple user terminals, and cross-app services. Zebra Smart also showcased the AutoClaw 2.0 in-vehicle solution, developed with data company YoooClaw, to bridge user data inside and outside the car. CTO Si Luo demonstrated always-on wake-free voice, active welcome, and seamless parking payment features, emphasizing that intelligent vehicles serve as the largest pilot testing ground for embodied AI. The company reports that 63% of enterprises deploying edge models in vehicles chose Zebra Smart's AutoOmni, which now partners with 10 chip companies globally.
Read sourceBanma Intelligence Launches AutoOmni 2.0 Edge AI Model for Smarter In-Car Experience
During the Yunqi Conference on September 23, Banma Intelligence (斑马智能) released its new full-modal edge-side large model, AutoOmni 2.0-23B-A3B, claiming task processing capabilities comparable to cloud models with 10 times the parameter count. The company also unveiled the AutoClaw 2.0 in-cabin collaboration service solution, aiming to make AI better understand the user's 'my world.' Banma Intelligence co-CEO Hao Fei stated that the core value of edge intelligence is enabling a new interaction paradigm through full-vehicle, always-on, wake-free interaction with context and memory. Chief Product Officer Cai Ming emphasized that AI must understand both the physical world (表世界) and the user's personal world (里世界) to achieve deep usage. The company reported that its 'Yuanshen AI' (元神AI) now covers 68% of mainstream automakers' services. AutoOmni 2.0 uses a Mixture of Experts (MoE) architecture and supports privacy-preserving, low-latency execution of complex tasks including navigation, vehicle control, and cross-domain operations. Banma Intelligence also announced partnerships with 10 chip companies and a security framework for automotive-grade edge AI. The company positions smart vehicles as the largest pilot field for embodied intelligence, and plans to evolve from AutoOmni to EdgeOmni, expanding from single-agent to group intelligence.
Banma Intelligence Launches AutoOmni 2.0 On-Device AI Model for Smart Cockpits
At the Yunqi Conference on September 23, Banma Intelligence (斑马智能) launched its next-generation full-modal on-device large model, AutoOmni 2.0-23B-A3B. According to the company, the model's task processing capability is comparable to cloud models with 10 times the parameter count. Banma Intelligence co-CEO Hao Fei stated that the core value of on-device intelligence is enabling a new interaction paradigm, with full-vehicle, always-on wake-free operation and contextual memory as key elements for proactive services. The model uses a Mixture of Experts (MoE) architecture. In smart cockpit scenarios, it handles ordinary tasks with capability equivalent to a 10x-parameter cloud model and complex tasks at 80-90% of that level. The company demonstrated a real-vehicle solution featuring the model and an end-cloud integrated agent architecture, capable of executing complex navigation, vehicle control, and multimedia tasks with low latency and privacy protection. Chief Technology Officer Si Luo noted that AutoOmni has partnered with 10 chip companies, supports 8+ concurrent tasks, achieves 5-6x inference acceleration, and includes a vehicle-grade on-device AI security system. The company claims 63% of enterprises deploying on-device models in vehicles chose Banma Intelligence in the past year.
Read source