Meta's New Free AI Can Watch, Listen, and Code for You

Meta just released Muse Spark Open, a new free AI model that can handle text, images, audio, and video all at once. Unlike many AI models that are kept secret, this one is open-weight, meaning anyone can download it and run it on their own computers. It's designed to be really good at coding tasks, like reading through a bunch of code files, finding bugs, and even watching a screen recording to understand what's going on. The model is massive—816 billion parameters—but it uses a clever trick called mixture-of-experts to only use a small part of itself for each task, so it runs faster and cheaper than you'd expect.

So what does this mean for you? If you're a developer, this could be a powerful tool that helps you write and debug code more quickly, and it's free to use. Even if you're not a coder, this shows that AI is getting better at understanding different types of information at the same time, which could lead to smarter assistants in the future. For example, you could show it a screenshot and ask a question, or give it a voice note and have it respond. The fact that it's open means anyone can build on it, so we might see new apps and services that use this technology soon.

The catch is that this model is huge—it needs about 1.5 terabytes of memory to run in full precision, which is way more than a typical laptop has. You'd need a powerful server with multiple high-end GPUs. However, there's a compressed version that uses 4-bit precision and runs on a single NVIDIA B200 GPU, which is still expensive but more accessible. Also, while it's great at coding, it might not be as good at other tasks as some closed models. But for developers and companies who want to customize their own AI, this is a big step forward.