Meta just changed sides.
For years, Meta’s AI strategy rested on a clear claim: open weights are good for the ecosystem, good for developers, and good for Meta’s competitive position against closed-model rivals. Llama was the embodiment of that claim. Muse Spark is its replacement, and it’s proprietary.
Forbes reported the model as Meta’s rebuilt AI stack following what the publication characterized as widespread criticism of the Llama 4 series. VentureBeat headlined its coverage “Goodbye, Llama?”, a framing that captures the market’s read of the strategic signal, even if the final chapter of that story hasn’t been written.
What Muse Spark actually is, according to Meta’s announcement: a multimodal model designed specifically for wearable deployment, optimized to process live camera and audio inputs from smart glasses in real time. It is not a general-purpose API model available to third-party developers. It operates within Meta’s own ecosystem. Meta describes it as performing competitively on health and multimodal benchmarks, with noted gaps in coding tasks, per the company’s internal evaluation, not an independent benchmark.
The acquisition context matters here. Meta reportedly acquired a 49% stake in Scale AI for approximately $14.3B, according to reports, a figure not confirmed in primary filings at the time of this brief. Alexandr Wang, Scale AI’s founder, reportedly joined Meta as Chief AI Officer following the acquisition. If accurate, this represents a structural bet: Meta is not just building a new model, it’s internalizing the data labeling and evaluation infrastructure needed to compete at the frontier without the open-source community it previously relied on for feedback and improvement.
The development timeline adds context. Muse Spark was reportedly developed over approximately nine months, a compressed cycle that suggests Meta treated this as an emergency response to Llama 4’s reception, not a deliberate multi-year strategic pivot.
For developers currently building on Llama, the practical question is what comes next. Muse Spark isn’t a replacement they can access, it’s a signal that Meta’s open-source commitment was conditional on competitive circumstances that no longer hold. The Llama series may continue in some form, but it’s no longer the company’s flagship investment. Practitioners should be auditing their dependency on Meta-maintained open weights now, not after a formal deprecation announcement.
The wearable angle is real, but secondary to the strategic story this cycle. Smart glasses as an AI interface category is worth tracking, it changes the interaction model from deliberate query to ambient, always-on processing. That has product implications, regulatory implications around privacy and consent, and a different set of performance requirements than a chat interface. Meta is betting this interface wins. Most of its competitors are still optimizing for the keyboard.
What this week’s launch makes clear: the era of major labs treating open-source release as a competitive strategy appears to be closing. Meta was the last large US lab publicly committed to open weights as a flagship model strategy. That commitment is now conditional at best.