Alibaba's Free AI Now Sees and Hears You in Real Time

Alibaba's Qwen team just released Voice and Video Chat, an AI you can talk to while showing it things on camera. It runs on Omni-7B, an "open-source" model — meaning the recipe is public and anyone can download it, free. "Multimodal" simply means one system handles text, sound, images and video together instead of stapling separate tools together. So instead of typing questions, you could hold a conversation, point your phone at a broken appliance, and get an answer out loud, in real time.

Why should you care? Almost all the good voice-and-video AI today — ChatGPT's voice mode, Google's Gemini Live — lives only on company servers. You don't own it, and your audio and video get sent to them. Qwen's model can be downloaded and run on your own machine, like a gaming PC or a rented cloud computer, which means private conversations could stay private. It's also free for developers, so expect a wave of cheaper voice assistants in customer service, translation, tutoring and accessibility tools.

The catch is quality and practicality. Open weights don't mean equal performance — the biggest paid models still tend to be smarter and smoother. Running video AI yourself takes a decent graphics card and some setup know-how, so most people will wait for someone to package it into an app. And Qwen is owned by China's Alibaba, which businesses handling sensitive data will weigh carefully. Real-time video AI also raises familiar worries about always-on cameras and fakes.

The bigger story is the trend: capable voice-and-vision AI is becoming free and copyable, fast. That pushes prices down and gives you more privacy options. Within a year, talking to your computer and showing it things may feel as ordinary as typing — and cost less than today's subscriptions.