Meta has rolled out Muse Spark, its first major AI model from the secretive TBD Lab led by 28‑year‑old Alexandr Wang. Hired after a string of setbacks, Wang was given $15 billion in funding and autonomy to rebuild the company’s research unit. In less than a year he assembled a team of roughly 100 top‑tier researchers and launched the model internally.

Supporters point to the model’s visual‑understanding capabilities as proof that Meta’s AI revival is gaining traction, and they expect follow‑up models later this year to narrow the gap with OpenAI, Google and Anthropic. Critics, however, argue Muse Spark was cobbled from existing Llama 4 code and datasets, labeling the progress incremental and warning that rival labs are moving faster.

Wang’s leadership style—startup‑like boba‑fuelled happy hours, flat hierarchies and a push for proprietary models—has reshaped Meta’s internal politics and safety work, sparking tension with legacy teams. The company plans to embed Muse Spark in its advertising and content‑ranking systems, while limited external API access makes performance hard to verify. Meta now banks on the next generation of models to justify its multibillion‑dollar AI spend.