Om AI (联汇科技)
Category: Foundation Models / LLMs
Chinese AI company building edge-side streaming multimodal models (VLX) for physical-world robotics, drones, and smart terminals, with a focus on end-side real-time perception-action pipelines. Om AI (联汇科技) was founded in 2003. The company is led by 赵天成 (Zhao Tiancheng). Based in Hangzhou, Zhejiang, China. Team size: 101-500. Total funding raised: $50.0M. Latest round: Strategic Investment. Key investors include China Mobile Industry Development Fund (中移和创投资), Qianhai Fund of Funds (前海方舟/前海母基金) via Zhongyuan Qianhai Fund and Qilu Qianhai Fund, ZhenFund (真格基金), Comprehensive Capital (综合资本).
- Founded
- 2003
- Headquarters
- Hangzhou, Zhejiang, China
- Team size
- 101-500
- Total funding
- $50.0M
Value proposition
Delivering the world's first end-side streaming multimodal model suite (VLX) purpose-built for edge devices, enabling real-time perception, precise grounding, and action decision-making on robots, drones, and smart terminals — designed from Day 1 for edge constraints rather than compressed from cloud models.
Products and solutions
VLX model series (VLX-Flow streaming perception, VLX-Seek precise grounding, VLX-Go action decision), VLM-R1 (first visual language model applying DeepSeek R1 reinforcement learning paradigm), OmModel multimodal LLM, OmBot autonomous AI agent, OmAgent developer community, AaaS (Agent-as-a-Service) platform, OttoBox AI Studio
Unique value
First company to build a streaming multimodal model natively for edge devices (not compressed from cloud), achieving the complete perception→grounding→action pipeline on-device with sub-0.06s latency for single video stream processing and 0.6B-parameter models outperforming larger alternatives.
Target customer
Enterprise B2B customers in telecom (China Mobile, China Telecom), energy/power (State Grid), media (CCTV/CNR), and smart terminal/robotics manufacturers; also B2B2C via telecom operators
Industries served
Telecommunications, Energy & Power (State Grid), Media & Broadcasting, Low-altitude Economy (drones), Robotics, Smart Terminals
Technology advantage
Day-1 edge-native architecture (not cloud-to-edge compression); Linear Attention with dual-layer memory for streaming video; Region Token approach replacing coordinate prediction for grounding; short-horizon waypoint prediction for action; shared base model across Flow/Seek/Go; first to apply DeepSeek R1 RL paradigm to VLMs (VLM-R1, 6000+ GitHub stars); 50+ patents and 50+ top conference papers
How they differentiate
Unlike competitors focused on large cloud-based multimodal models, Om AI designs models natively for edge deployment with streaming real-time perception, grounding, and action capabilities. They prioritize understanding over generation, have a unique B2B revenue model serving state-owned enterprises, and leverage a 20-year legacy 2B business foundation (联汇科技) for distribution and customer trust.
Main competitors
SenseTime (商汤科技), StepFun (阶跃星辰), Baidu (ERNIE multimodal)
Key partnerships
China Mobile (strategic investment + joint AI solutions), China Telecom, State Grid, CCTV/CNR (China Media Group), Zhejiang University (Om AI Research Center), Huawei (Ascend computing partnership), Alibaba Cloud
Notable customers
China Mobile, China Telecom, State Grid Corporation of China, China Media Group (CCTV/CNR)
Major milestones
2020: Founded by Zhao Tiancheng, 2021: Released first self-developed multimodal model, 2022: First company in China to receive national工信部 multimodal model certification, 2023: Released OmModel V3.0 and OmBot autonomous agent, 2024: Completed strategic funding round led by China Mobile, 2025: VLM-R1 open-source release (6000+ GitHub stars), 2026: Launched VLX streaming multimodal model series at CVPR 2026
Growth metrics
VLM-R1 achieved 6000+ GitHub stars, #1 on GitHub global trending within 48 hours; VLX launched at CVPR 2026; revenue-generating with major state-owned enterprise customers
Market positioning
Leading Chinese edge-side multimodal AI company; recognized as a "quasi-unicorn" (准独角兽) in Hangzhou; one of the earliest Chinese companies to receive national certification for multimodal models (2022); positioned as the "brain" for physical-world AI terminals
Geographic focus
China (primary); expanding via joint ventures (e.g., Xinjiang production and construction corps JV)
Patents and IP
50+ invention patents; 50+ top conference papers at CVPR, ACL, ECCV, AAAI; National Science and Technology Progress Award; Wu Wenjun AI Science & Technology Progress Award winner
About 赵天成 (Zhao Tiancheng)
Ex-CMU LTI PhD; UCLA BS Electrical Engineering. CMU Computer Science PhD (4.5 years). Published 20+ top-tier conference papers. Co-founder of 联汇科技 (Lianhui Technology), a 20-year 2B enterprise founded by his father Zhao Fan.
Latest news about Om AI (联汇科技)
More Foundation Models / LLMs companies
Official website: https://www.hzlh.com
