The landscape of open-source artificial intelligence has long been populated by models like Mistral and Falcon, offering developers accessible options. Meta’s entry into this space with its Llama series represented a significant shift, leveraging its vast user base and resources to champion open-weight models. This approach resonated deeply, with the Llama ecosystem reportedly reaching 1.2 billion downloads by early 2026.

A New Proprietary Direction

This context makes Meta’s announcement on April 8, 2026, particularly notable. The company launched Muse Spark, the first major new model from its Meta Superintelligence Labs. While benchmarks show it as a capable competitor against current frontier models, its release strategy marks a definitive departure. Unlike its predecessors, Muse Spark is entirely proprietary, with no free download or open weights available.

The development followed a substantial internal overhaul. Meta invested $14.3 billion and appointed Alexandr Wang, formerly of Scale AI, to lead a complete rebuild of its AI infrastructure. After nine months of work, Muse Spark emerged as the result, a natively multimodal reasoning model with integrated tool-use and multi-agent orchestration.

Capabilities and Performance

This new model now powers the Meta AI assistant across the company’s apps, reaching billions of users. The rebuilt infrastructure allowed Meta to create a model with capabilities comparable to its older midsize Llama 4 variant but at a significantly reduced computational cost. This efficiency is critical for deploying AI at Meta’s global scale.

On standardized benchmarks, Muse Spark’s performance is varied. It scores 52 on the Artificial Intelligence Index v4.0, placing it fourth overall behind leading models from Google, OpenAI, and Anthropic. Meta has notably avoided claiming it has built the world’s best model, a contrast to earlier promotional strategies.

Where Muse Spark demonstrates a clear lead is in the health domain. On the HealthBench Hard evaluation for open-ended health queries, it scores 42.8, substantially ahead of its closest competitors. Meta states that health is a priority and that it collaborated with over 1,000 physicians to curate training data for this purpose.

The model offers three interaction modes: Instant for quick answers, Thinking for multi-step reasoning, and Contemplating, which orchestrates multiple agents in parallel to compete with the most advanced reasoning modes from other companies.

Developer Community Reaction

The move away from open-source principles is a central part of the Muse Spark story. The developer community that propelled Llama’s adoption is now presented with a closed system. Meta has stated it will offer the model in a private preview via an API to select partners, making it more restricted than some rival paid offerings.

Alexandr Wang addressed the strategic change directly, explaining that the rebuild was step one and that bigger models are in development with plans to open-source future versions. The response from developers has been skeptical. Some view it as a necessary pivot, while others interpret it as Meta protecting valuable technology after establishing market presence.

Distribution and Implications

Meta is not delaying its rollout for developer approval. Muse Spark is set to debut within weeks across Facebook, Instagram, WhatsApp, Messenger, and Meta’s Ray-Ban smart glasses. This direct deployment to over three billion existing users represents a distribution advantage distinct from companies that primarily sell to enterprises.

Meta’s focused push into health AI with Muse Spark inevitably raises privacy considerations. Users must log in with a Meta account, and while the company has not explicitly stated that personal account data trains the AI, it has historically trained on public user data and positions Muse Spark as a personal superintelligence product.

The market reacted positively to the launch, with Meta’s stock rising more than 9% on the announcement day, indicating investor confidence in the new strategic direction.

Looking ahead, the industry will monitor two key trajectories: the integration and user adoption of Muse Spark across Meta’s ecosystem, and the company’s adherence to its stated, but unscheduled, plan to return to open-source releases with future model generations. The success of this proprietary phase will likely determine the timeline and nature of that promised openness.