Today's signal
Meta launched Muse Spark, the first AI model from Alexandr Wang's Meta Superintelligence Labs, built from scratch in nine months. It's live now on the Meta AI app in the US, with WhatsApp, Instagram, Facebook, and Messenger rollout coming in weeks. The Meta AI app jumped from #57 to #5 on the US App Store within 24 hours.
Why it matters
Zuckerberg spent $14.3 billion to bring in Wang after Llama 4 flopped. And this is his first deliverable. The benchmarks are legitimately strong: Muse Spark leads on health reasoning (42.8 on HealthBench Hard vs GPT-5.4's 40.1), tops visual understanding (86.4 on CharXiv), and sits in frontier range on PhD-level reasoning (89.5 on GPQA Diamond). Gaps in coding and abstract reasoning remain, and Meta acknowledges them. But no other AI lab has what Meta has: a path to 3 billion users, and a model that's free — no tiers, no paywall — while every competitor is raising prices.
The take
Everyone counted Meta out. That was premature. Muse Spark doesn't need to be the best model. It needs to be good enough inside apps that people already open 30 times a day. OpenAI has ~400 million users. Meta is about to reach 7x that. Wang delivered in nine months, future Muse models will be open-sourced, and the timing is perfect: competitors are building paywalls while Meta is tearing them down. Distribution wins.
The number
10x — the compute efficiency gain. Meta says Muse Spark matches Llama 4 Maverick's capabilities with over 10x less compute. That's not just a benchmark story — it's what makes it possible to serve a free model to 3 billion users without bankrupting the ad business.
👉 Muse Spark: Alexandr Wang's First Output Puts Meta Back in the AI Race — full breakdown on Analytics Drift