Realtime AI News
Banma Smart Releases On-Device Omni-Modal Model AutoOmni 2.0 for Its Yuanshen AI
Banma Smart introduced AutoOmni 2.0-23B-A3B, a new generation on-device omni-modal large model, during the Yunqi Conference on September 23. The company frames the release as making its Yuanshen AI understand "my world" better.
Banma Smart introduced AutoOmni 2.0-23B-A3B on September 23 during the Yunqi Conference, describing it as a new generation of on-device omni-modal large model.
Two words in the release carry most of the weight: omni-modal and on-device. Omni-modal means the model handles several kinds of input rather than text alone, while on-device means inference runs locally on the device instead of relying entirely on the cloud.
The model designation is AutoOmni 2.0-23B-A3B. The 23B portion refers to model scale, and the A3B suffix is typically used to distinguish the portion active during computation. Banma Smart did not elaborate on the parameter breakdown, the supported modalities, target hardware, or benchmark results in the release information.
The appeal of on-device multimodal models is mostly about latency, privacy, and availability offline. Keeping interpretation on the local device cuts round trips to the cloud and preserves basic functions when connectivity is poor, which matters in vehicles and wearables.
Yuanshen AI is where the release lands. By tying the model to an agent experience rather than shipping weights alone, the company frames AutoOmni 2.0 as an upgrade to something users already interact with.
The Yunqi Conference supplies the timing and the setting. Announcing during a major industry gathering usually signals that more device and scenario details are coming, since hardware partners and developers are in the same room.
What to watch: which devices get AutoOmni 2.0 first, how demanding the local hardware requirements are, and whether "understanding my world better" can be measured in real tasks rather than asserted.
Why it matters
Moving omni-modal understanding onto the device lowers latency and reduces dependence on sending data to the cloud, shifting competition toward local compute and hardware integration.
Nearby Updates
All09/23, 16:54
Report: Meta Is Testing Human Contractors to Handle Calls for Its Muse AI Agent
Analytics India Magazine reports that Meta is testing human contractors to take over calls for its Muse AI agent, filling gaps the agent cannot handle on its own. The story is attributed to a report rather than an official announcement, and the scale, regions, and launch timing remain undisclosed.
09/23, 15:27
Qwen Names Liu Dayiheng as Its New Lead
Qwen has settled who runs the model family, according to a report from QbitAI, with Liu Dayiheng taking over the top role. The report notes he comes from the same cohort of “talent prodigy” engineers as the widely known ZhiHuiJun, and the change could affect release cadence, open-source strategy and community communication.
09/23, 13:16
Zhi Huijun prices a robot at 20,000 yuan per unit — humanoid, quadruped, and open to developers
According to QbitAI, Zhi Huijun is offering a robot priced at 20,000 yuan per unit, one that can take a humanoid or a quadruped form and is open to developers. The report frames the price with a direct question: what kind of robot do you actually get for 20,000 yuan?
09/23, 13:13
Alibaba's Qwen AI platform gets a full upgrade across model, agent, and application services
QbitAI reports that Alibaba's Qwen AI platform has been upgraded across model services, agent services, and AI applications. The update spans MaaS, agent tooling and industry-specific AI solutions, covering the path from model calls through to business deployment.