Meta's New AI Model Is Huge But Not the Smartest
Bigger isn't always better: Meta's 816B model ties with a 180B model.
Meta has released a new open AI model called Muse Spark Open, and it's enormous—816 billion parameters. Parameters are like the model's brain cells; more usually means smarter. But when tested by Artificial Analysis, which ranks AI models, it scored 39 out of 100. Surprisingly, a model less than a quarter its size, Qwen3.8-Flash-Next with 180 billion parameters, scored 40. That's a win for efficiency: smaller models can match giants.
So what? If you're a business or developer, you might not need to spend big on hardware. Meta's model requires two high-end Mac Studios with 512GB memory each to run locally, while the smaller Qwen model needs far less. For everyday users, this means AI tools could become cheaper and faster without sacrificing quality. The smaller model also uses less energy, which is better for the environment.
But there's a catch: the Artificial Analysis score is just one number. It doesn't measure speed, how well the model understands long documents, or whether it can analyze images. It also doesn't predict how the model will perform on your specific tasks, like writing emails or answering customer questions. So while the score is a useful starting point, real-world testing is still key.
Other models in the chart show even higher scores, like MiMo-V2.6-Pro at 46, but they come with their own trade-offs. The takeaway? The AI race isn't just about size anymore—it's about getting the most value for your resources. As a user, you benefit from more choices and potentially lower costs.
- Meta's 816B model scores 39, while Qwen's 180B model scores 40 on a quality test.
- Smaller models can be cheaper and easier to run, saving money and energy.
- The score doesn't cover speed, vision, or real-world task performance.
Why It Matters
You might get equally good AI from smaller, cheaper models, saving money and energy.