资本重仓端侧物理AI:前海母基金数亿元押注,Om AI联汇加速端侧AI商业化落地
TL;DR - Hangzhou-based Om AI (联汇) closed a several-hundred-million-RMB round led by Qianhai FOF with Hangzhou government industrial fund participation, and simultaneously open-sourced VLX-Seek 1.5, an "edge-native" streaming fine-grained perception multimodal model for physical AI. It matters as a signal that capital and deployment focus is shifting from cloud training scale to on-device perception for embodied/physical AI.
- VLX-Seek 1.5 is positioned as an edge-native streaming multimodal perception model built on the company's in-house VLX foundation, closing a loop of "continuous perception → precise localization → action decision."
- The 3B variant is claimed to beat NVIDIA's LocateAnything-3B at equal parameter scale, with +62.9% accuracy in drone embodied scenarios and −74.8% false-positive rate on targets (vendor-reported figures, not independently verified).
- Commercialization spans a "one brain, many bodies" strategy: OttoBox AI Studio for Lenovo/Apple AIPC hardware, partnerships with embodied-intelligence firms, and the Homer AI wearable vision aid serving ~100k visually impaired users in China at ~10M monthly AI calls.
- Note: this is a company-supplied press release republished by 量子位, so benchmark claims and funding details come from the vendor rather than third-party reporting.