Meta goes proprietary with Muse
Meta launched Muse Spark, its first major model from the Superintelligence Labs and a clear shift toward a proprietary stack rather than the company’s earlier open-source posture. (artificialintelligence-news.com) The launch drove a sharp day‑over‑day spike in downloads for the Meta AI app, but critics point to past Llama 4 benchmark issues and underperformance as reasons the company doubled down on a closed approach. (newsbytesapp.com) (futurism.com)
Meta spent years telling developers that open models were the future, and then on April 8 it introduced Muse Spark as a model that powers the Meta AI app and website without releasing the weights to the public. Meta said Muse Spark is the first model in a new Muse series built by Meta Superintelligence Labs after a nine-month rebuild of its artificial intelligence stack. (about.fb.com) That is a real turn for a company that made Llama into its flagship open-family brand. Muse Spark is the first major release tied to the lab Meta formed around former Scale AI chief executive Alexandr Wang after Meta bought a 49 percent stake in Scale AI for $14.3 billion. (techcrunch.com) Meta is pitching the new model less like a research gift and more like a product engine. The company said Muse Spark already runs Meta AI on the web and in the app, and will roll out to WhatsApp, Instagram, Facebook, Messenger, and Meta’s artificial intelligence glasses in the coming weeks. (about.fb.com) The immediate payoff showed up in app charts, not in open-source repositories. TechCrunch reported that the Meta AI app jumped from No. 57 in the United States Apple App Store before the launch to No. 5 the next day, while Sensor Tower estimated about 46,000 United States iPhone downloads on April 8, up 87 percent day over day. (techcrunch.com) That jump helps explain why Meta may prefer a closed model here. If the best version only lives inside Meta’s own apps, every improvement can pull people into products Meta controls instead of giving outside developers the same engine for free. (about.fb.com) (techcrunch.com) The backdrop is that Llama stopped looking like an easy public-relations win. Meta’s April 2025 Llama 4 release drew criticism over benchmark presentation and over whether the models matched the hype in real use, which weakened the case for winning mindshare by publishing model weights first and asking product questions later. (futurism.com) Meta’s own language around Muse Spark also sounds different from the old Llama era. The company says the Muse line will scale in steps, with each generation validating the last before it goes bigger, which is the language of an internal product roadmap more than a broad open ecosystem. (about.fb.com) There is still a second act coming. TechCrunch reported that Meta plans a “Contemplating” mode for harder problems, which suggests Muse Spark is being launched as a live consumer service that Meta can tune in public instead of a static model drop that the outside world forks on day one. (techcrunch.com) So the story is not just that Meta shipped a new model on April 8. It is that the company appears to be moving from “here are the weights” to “come use our app,” and the first evidence says that strategy can move millions of users faster than another open release could. (about.fb.com) (techcrunch.com)
