Meta is no longer experimenting with AI. It is restructuring around it.

Meta has launched Muse Spark, the first AI model from its new Superintelligence Labs, marking an early test of how the company plans to compete more directly with OpenAI, Google, and Anthropic.

Why You Should Care

This is the first real signal of how Meta intends to compete in the next phase of the AI race.

Muse Spark is not just a new model. It reflects a shift in strategy. Meta is moving from open experimentation toward a more controlled and potentially monetizable AI stack. For operators, this changes how Meta fits into the broader ecosystem, not just as a platform, but as a direct infrastructure and product competitor.

The tech giant introduced Muse Spark as the first model developed by its new Superintelligence Labs, led by Chief AI Officer Alexandr Wang. The model will power the Meta AI chatbot across the company’s apps, including Instagram, Facebook, and WhatsApp.

Unlike Meta’s earlier Llama models, Muse Spark is closed, meaning its architecture and code will not be publicly released. This marks a clear departure from the company’s previous open-source positioning and gives Meta more control over how the model is distributed and integrated into products.

The model was built over nine months and is considered internally as an early step in a broader pipeline of AI systems, with larger models already in development. The company has also indicated it may offer API access to the model in the future, while keeping its chatbot free for users for now.

Muse Spark supports multiple reasoning modes, including Instant, Thinking, and Contemplating, with strengths in areas such as science, health, and math. It is currently less competitive in coding compared to leading models from OpenAI, Google, and Anthropic.

Beyond chat, the model is already being applied in early product use cases, including a shopping assistant designed to improve product discovery across Meta’s platforms.

The Ripple

The shift from open to closed models is not just a Meta story. It reflects a broader recalibration across the AI industry.

As competition intensifies, companies are balancing openness with control. Closed models allow tighter product integration and clearer monetization paths, especially as AI becomes embedded into large-scale consumer platforms. Meta’s move signals that even companies that helped push open-source AI are now adjusting to a more competitive and commercially driven environment.

There is also a geopolitical layer. Muse Spark was trained using a mix of third-party models, including Qwen from Alibaba, alongside models from US companies. This highlights how interconnected the global AI stack remains, even as policymakers and companies push for tighter control over model development and capabilities.

What to Watch

The key question is not whether Muse Spark outperforms current leading models.

It is whether Meta can build a consistent release cycle and product layer around it.

If Muse Spark becomes the first in a sequence of rapidly improving models tied closely to Meta’s distribution, the company could shift from being a fast follower to a more structured competitor in AI. The model itself is an early step. The cadence that follows is what matters.