Our Take
This is a seismic shift. Meta going proprietary after building its brand on open-source means the economics of frontier AI have changed. 10x compute efficiency over Maverick is the kind of gain that justifies the pivot.
The Pivot
Just weeks after releasing Llama 4 Scout and Maverick as open-weight models, Meta reversed course entirely with Muse Spark — its first-ever proprietary AI model. Developed by Meta Superintelligence Labs under Alexandr Wang (acquired in a $14 billion deal), Muse Spark signals a fundamental shift in Meta's AI strategy.
Performance
Meta reports that Muse Spark achieves its reasoning capabilities using over an order of magnitude less compute than Llama 4 Maverick. It scored 52 on the Artificial Analysis Intelligence Index, compared to Maverick's 18 — a dramatic efficiency improvement.
Llama 4 Still Available
Llama 4 Scout, a 17-billion-parameter vision-language model optimized for edge devices, remains open-source and runs on a single consumer GPU with 24GB VRAM. Meta is continuing to develop both open and proprietary models.
LlamaCon
Meta will share more about its dual-track AI strategy at LlamaCon on April 29, 2026.