On April 8, 2026, Meta announced Muse Spark (code-named Avocado internally), the first flagship AI model from its newly formed Meta Superintelligence Labs—a $14.3 billion bet on 28-year-old Alexandr Wang's vision for artificial intelligence dominance.
The launch represents a seismic shift in Meta's AI strategy. For years, the company positioned itself as the open-source champion, freely distributing its Llama models to researchers and developers worldwide. Muse Spark breaks that legacy entirely. It's a closed, proprietary model available only through Meta's own products and a select group of API partners—a dramatic reversal that underscores how seriously the company is taking the AI race against OpenAI, Google, and Anthropic.
The Wang Effect
Wang's arrival at Meta last June came with an extraordinary mandate: rebuild the entire AI stack from the ground up. As Chief AI Officer—a position that didn't exist before his hire—Wang has been given unprecedented autonomy to reshape Meta's approach to artificial intelligence. The company's $14.3 billion acquisition of a 49% stake in Scale AI was essentially an acquisition of Wang himself, signaling Zuckerberg's belief that talent and vision matter more than incremental improvements to existing systems.
The shift is palpable. Where Meta's previous AI leadership emphasized openness and democratization, Wang's superintelligence labs operate with a different philosophy: build the best, keep it proprietary, and monetize through Meta's massive user base of 3+ billion people.
What Is Muse Spark?
Muse Spark is positioned as a multipurpose AI model that accepts voice, text, and image inputs while producing text-only outputs. The model features three distinct operational modes:
- **Quick Answer Mode**: For straightforward queries requiring immediate responses - **Complex Task Mode**: For intricate problems requiring deeper reasoning - **Contemplating Mode**: A rolling rollout feature enabling extended thinking and analysis
According to Meta's internal benchmarks, Muse Spark is competitive with leading frontier models from OpenAI, Anthropic, and Google across many tasks, though it doesn't uniformly surpass them across all domains. The company highlights particular strength in medical and scientific reasoning—areas where accurate, nuanced responses are critical.
Muse Spark will initially power Meta AI queries in the Meta.ai website and Meta AI app, with broader rollout planned across Facebook, Instagram, WhatsApp, and Messenger in the coming weeks. The model will also power queries in Meta's Ray-Ban smart glasses, extending AI capabilities into wearable devices.
“We are building products that don't just answer your questions but act as agents that do things for you.”— Mark Zuckerberg, Meta CEO
The Closed-Model Gamble
The decision to keep Muse Spark proprietary is perhaps the most controversial aspect of this launch. Meta built its AI reputation on Llama—a family of open models that spawned countless research projects, startups, and innovations. By closing Muse Spark, Meta is explicitly choosing user reach and defensibility over developer goodwill.
The company did leave a door open to future openness. Meta stated it hopes to eventually open-source future versions of Muse Spark, though no timeline was provided. This suggests the company views proprietary closure as a temporary competitive advantage, not a permanent strategy shift.
Mark Zuckerberg framed the vision in a statement on Threads: "We are building products that don't just answer your questions but act as agents that do things for you." He added that Meta's goal is to deliver "personal superintelligence in everyone's hands." This language signals that Muse Spark is just the beginning of a broader push toward AI agents that can autonomously handle tasks like shopping, travel planning, and communication.
The Investment Case
Meta's commitment to AI scale is staggering. The company projects AI-related capital expenditure between $115 billion and $135 billion in 2026 alone—nearly double its 2025 spending. That's roughly equivalent to the entire annual R&D budget of most Fortune 500 companies, devoted almost exclusively to a single technology.
For Zuckerberg, this isn't reckless spending—it's existential. Meta's core advertising business faces pressure from regulatory scrutiny and privacy changes. The company sees AI as both a new growth vector and a hedge against future disruption. If Muse Spark succeeds in driving engagement and enabling new use cases across Meta's platforms, the returns could justify the enormous capital commitment.
What's Next
The arrival of Muse Spark doesn't mean Meta's AI journey is complete. It's the opening salvo in what Wang and Zuckerberg clearly envision as a years-long competition for AI leadership. The Contemplating mode rollout, still in gradual deployment, will eventually enable more sophisticated reasoning. Future versions may bring new capabilities or address current limitations.
For the broader AI industry, Muse Spark's success or failure matters. If Meta can leverage its unmatched distribution advantage to achieve Muse Spark adoption at scale, it could reshape competitive dynamics. OpenAI's ChatGPT won partly through accessibility and word-of-mouth. Meta has 3+ billion users already. If Muse Spark proves compelling enough that people use it on WhatsApp, Instagram, and Ray-Bans without actively seeking it out, Meta could rapidly capture mind share in ways that pure-play AI companies cannot.
The closed-model strategy is a bet that distribution trumps openness in the AI era. If Wang and Zuckerberg are right, Meta just announced the model that could define the next decade of AI.