Meta unveils Muse Spark, a multimodal model pitched as the first rung toward “personal superintelligence”

What’s new

Meta says it has launched Muse Spark, a natively multimodal reasoning model from its Meta Superintelligence Labs, and it’s available today at meta.ai and in the Meta AI app. The company frames the release as a “ground-up overhaul” of its AI efforts — a fresh scaling ladder built from research through to new infrastructure like the Hyperion data center. It has been reported that a private API preview will open to a limited set of users.

Muse Spark is billed to support tool use, visual chain-of-thought, and multi-agent orchestration. Contemplating mode, a key feature, allegedly runs several reasoning agents in parallel and is positioned to compete with extreme-reasoning modes in frontier models such as Gemini Deep Think and GPT Pro. Meta reports performance gains on internal benchmarks — it has been reported that Muse Spark achieved 58% on “Humanity’s Last Exam” and 38% on “FrontierScience Research” — details which the company says are explained in an accompanying methodology document.

Personal use cases — promising, human, and a bit uncanny

Meta emphasizes personal applications: visual troubleshooting of home appliances, interactive mini-games, health reasoning, and personalized fitness or yoga coaching. For health use-cases the company says it collaborated with over 1,000 physicians to curate training data so the model can generate more factual, interactive displays — it has been reported that this physician collaboration informed Muse Spark’s health reasoning. Example prompts in the announcement range from “turn this into a playable web sudoku” to “highlight coffee-machine components and show an interactive latte-making tutorial,” and even personalized diet overlays and side-by-side yoga form scoring with hoverable annotations. Neat, right?

But here’s the emotional pivot: this is about systems that not only answer questions but try to understand your world — your kitchen, your meals, your body. That’s exciting. And it’s uncomfortable. Who watches the watchmen? Meta positions Muse Spark as a helper for everyday tasks and wellness, yet these same capabilities raise familiar questions about privacy, safety, and evaluation transparency. Meta’s blog points readers to methodology docs and promises continued scale-up — we’ll see whether the reality matches the rhetoric.

Sources: meta.com, Hacker News