Meta's AI Crisis: Fudged Benchmarks, a $15B Hire, 15,000 Layoffs, and the Death of Fully Open Source

Meta's AI strategy is in upheaval. After Llama 4 launched with fudged benchmarks in April 2025 — confirmed by departing chief scientist Yann LeCun — CEO Mark Zuckerberg sidelined the entire GenAI organization and brought in Scale AI's Alexandr Wang via a $15 billion deal to lead a new Superintelligence Lab. Wang's first models (Avocado for text, Mango for multimedia) are delayed and trailing Google, OpenAI, and Anthropic in internal benchmarks. Meta is abandoning fully open source AI — the largest models will stay proprietary, with smaller versions open-sourced later. The company is spending $115-135 billion on AI infrastructure in 2026 while cutting 15,000 jobs. LeCun called Wang 'inexperienced' on his way out the door. Here is what happened, what's coming, and what it means for the open-source AI ecosystem.