From video comprehension to edge implementation, Om AI aims for practical AI solutions.

      Credit: BEYOND EXPO

      As competition among large models transitions from parameter size to practical deployment capability, several Chinese companies specializing in edge AI are attracting attention, including Om AI Technology.

      Established in 2021, Om AI has opted not to chase extremely large cloud-based models; instead, it concentrates on general-purpose multimodal vision models for edge devices, aiming to integrate AI into physical devices such as PCs, cameras, and robots.

      During the BEYOND Expo 2026 media day, Om AI Technology unveiled its AI-native content creation product, OttoBox AI Studio. Tailored for media professionals and content creators, this tool utilizes local AI computing capabilities to offer functions like video analysis, asset matching, script generation, and quick video production. The company promotes it as a creative partner for the AI-native era, designed to enhance creative efficiency.

      In contrast to many AI firms that evolve from general-purpose models to application layers, Om AI is rooted in industry-specific needs. The team has extensive experience in the media and audiovisual sectors, which informs their approach to developing models that solve real-world issues rather than merely seeking universal solutions.

      Dr. Zhao Tiancheng, the CEO of Om AI, highlighted that this long-term industry background not only expedites model deployment but also ensures access to vast amounts of high-quality real-world data. They argue that genuine multimodal capability involves comprehending video, audio, and text concurrently, rather than merely distinguishing among them.

      A primary technical focus for the company is video understanding via low-parameter models. Unlike traditional methods that depend on significantly larger parameter counts and cloud-based GPU resources, Om AI advocates for a compact, precise, and rapid edge-model strategy. By minimizing model size, their AI can operate directly on local devices, which reduces inference costs and data upload demands while addressing business concerns about data security and privacy.

      This advantage in edge deployment is especially crucial for large-scale video analysis applications. The company's models are reported to achieve millisecond-level inference speeds, making them ideal for real-time uses such as security monitoring, industrial inspection, and AIoT analytics.

      Currently, Om AI’s AI ventures encompass three primary domains: AI PCs, AIoT, and embodied intelligence. Along with partnerships with major brands like Apple, Lenovo, and HP, their models are implemented in robots, robotic dogs, and drones, endowing these devices with autonomous decision-making and action capabilities.

      Om AI is also delving into inclusive AI applications. For instance, its Homer App is designed for visually impaired users, facilitating object searches and assisted navigation via smartphones or AI glasses.

      The flagship version of OttoBox AI Studio has established strong partnerships with prominent PC manufacturers such as Apple, Lenovo, and HP, firmly positioning itself in the AI PC sector and offering professional users an immediate, ready-to-use experience.

      This year, the company’s primary strategic focus is to launch its next-generation edge multimodal model, VLX, which aims to further enhance video comprehension and decision-making while consistently lowering operational costs. As the AI industry shifts from cloud-centric competition to on-device deployment, companies like Om AI are poised to be pivotal in driving the real-world adoption of multimodal AI.

      Jessie Wu is a technology journalist based in Shanghai, covering consumer electronics, semiconductors, and the gaming sector for TechNode. You can reach her via email at jessie.wu@technode.com.