Meta Launches Muse Spark — Its First Proprietary AI Model

Meta just released Muse Spark, its first AI model built from the ground up by the company’s new Superintelligence Labs. It marks a significant shift in Meta’s AI strategy — and the biggest surprise is that it’s not open source.

What Is Muse Spark?

Muse Spark is a multimodal AI model that accepts text, voice, and image inputs. You can point your phone’s camera at something and ask Meta AI to identify it, or speak to it naturally through voice. It currently powers the Meta AI app and website and will roll out to WhatsApp, Instagram, Facebook, Messenger, and Meta’s Ray-Ban smart glasses in the coming weeks.

The model was built by Meta Superintelligence Labs, the division led by Alexandr Wang, who joined Meta nine months ago in a deal reportedly worth $14 billion. Internally codenamed “Avocado,” Muse Spark is the first model in Meta’s new Muse series and replaces Llama 4 as the engine behind Meta AI.

Three Ways to Think

Muse Spark offers three reasoning modes. Instant gives you a fast answer with no extra processing. Thinking adds a short pause where the model works through multi-step problems in math, analysis, or logic. Contemplating is the most powerful — it orchestrates multiple AI agents that reason in parallel, competing with deep reasoning modes from OpenAI’s GPT-5.4 and Google’s Gemini Deep Think.

How It Stacks Up

On the Artificial Analysis intelligence index, Muse Spark scored 52 — behind Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. But its standout is health: it scored 42.8 on HealthBench Hard, nearly triple Gemini’s 20.6 and far ahead of Claude’s 14.8.

Meta says improved training techniques let it build smaller models as capable as the older Llama 4 Maverick using “an order of magnitude less compute.”

The Open-Source Question

For years, Meta’s Llama models defined the open-source AI movement. Muse Spark breaks that pattern entirely. The model is proprietary, and only select partners can access a private API preview. Meta says it “hopes to open-source future versions,” but for now, this is closed.

What This Means for You

If you use Meta AI on any of Meta’s apps, you’ll notice a significant upgrade in the coming weeks — especially for image understanding, health questions, and complex reasoning tasks. If you want to try it now, the Meta AI app is available in the US.

For those building with AI, the shift away from open source is worth watching. If Meta keeps its best models proprietary, the open-weight landscape loses its biggest champion. For more on how to pick the right AI tool for your workflow, see our Claude complete guide or explore our getting started page.