Meta launches Muse Spark as Zuckerberg resets AI strategy

Meta Platforms is sharpening its focus on artificial intelligence, moving beyond early experiments as competition in the sector intensifies. 

The company has introduced Muse Spark, the first system from its new Superintelligence Labs, offering a clearer view of how it plans to take on rivals such as OpenAI, Google and Anthropic. 

For Mark Zuckerberg, the rollout signals a change in approach.

The company is putting more emphasis on control, product integration and revenue opportunities tied to AI across its apps. 

That shift means AI is no longer treated as a side project.

Instead, Meta is building it into the core of its services, aiming to strengthen its position both with users and developers as demand for generative AI tools grows.

A more controlled AI strategy 

Muse Spark sits at the center of this plan.

Built over nine months, it will power Meta AI features across Instagram, Facebook and WhatsApp, expanding the company’s reach in everyday digital interactions. 

In a notable shift, the model is not open-source.

That marks a break from Meta’s earlier Llama systems and gives the company tighter control over how the technology is used and distributed.

It also opens the door to future revenue through developer access, including possible APIs. 

Muse Spark includes different reasoning modes—Instant, Thinking and Contemplating—and performs well in areas such as science, health and mathematics.

It is still catching up in coding, where competing tools from OpenAI and Google remain ahead. 

Meta is also testing the model in commercial features, including shopping tools designed to help users find products more easily across its platforms.

Industry shifts and global ties 

Meta’s move reflects a wider change across the AI industry.

As competition heats up, companies are weighing the benefits of open systems against the need for tighter control and clearer business models. 

Closed systems can make it easier to connect AI tools directly to products and services, while also giving companies more control over how they earn from them. 

There is a global angle as well. Muse Spark was trained using a mix of systems, including Qwen from Alibaba Group, alongside U.S.-built models.

That mix shows how connected AI development remains, even as governments push for closer oversight.

What comes next 

The bigger test for Meta will be consistency.

The company needs to keep improving the model, release updates regularly and deepen its use across its apps. 

Muse Spark is an early step. What matters now is how quickly Meta can build on it and whether it can turn that progress into products people use every day.