Meta put out Muse Spark on Wednesday, the company’s first major artificial intelligence model in over a year under new AI chief Alexandr Wang.

Meta (NASDAQ: META) shares climbed as much as 9% after the announcement, erasing losses from late March as the social media company tries to catch up with AI leaders like OpenAI, Google, and Anthropic.

The AI model, previously called Avocado, is available on Meta’s AI website and app now. Meta described it as the first product of a complete overhaul of the company’s AI efforts. According to Meta, Muse Spark handles the same work as the earlier Llama 4 Maverick model but uses less computing power to do it.

This is Meta’s first AI release with Wang running things. Wang is a billionaire tech executive who came to Meta after the company invested $14.3 billion in his firm, Scale AI. Meta brought him on as chief AI officer and put him in charge of the Superintelligence Labs team.

Shift from open-source approach

Unlike Meta’s previous models, Muse Spark breaks from the company’s open-source tradition. The model is closed and proprietary, marking a sharp change from past strategy. The company will give private preview access to some partners through an API, but hasn’t committed to releasing the model publicly.

Meta says it hopes to open-source future versions.

The model’s release was delayed because early versions didn’t perform well enough in benchmark tests. Muse Spark failed to beat competing models from Google, OpenAI, and Anthropic. But Meta’s new comparison data shows Muse Spark now matches or beats rival AI systems in different performance measures. Independent testing shows the model does well on health-related questions but still needs work on coding tasks.

Meta expects to spend $135 billion on AI this year. That’s nearly double what it spent in 2025. The big spending comes after some problems in the AI race.

Meta’s Llama 4 model didn’t live up to expectations when it launched in 2025. The company abandoned development of its largest variant, codenamed Behemoth, which would have reached 2 trillion parameters.

After that setback, Meta bought a 49% non-voting stake in Wang’s Scale AI for about $14 billion last summer. Then Meta committed $600 billion toward AI infrastructure in the U.S. through 2028.

Job cuts follow Metaverse losses

The AI push meant cuts in other areas. Meta cut hundreds of jobs last year, including workers in its Reality Labs division. That division, which built the now-defunct Metaverse, lost $80 billion. Its Horizon Worlds platform, a social virtual reality system, had less than 200,000 monthly active users. The goal was 500,000.

Meta also announced a “Contemplating mode” for Muse Spark that will roll out gradually. This feature uses multiple reasoning agents working together to handle complex problems. The model includes what Meta calls natively multimodal reasoning with tool-use and visual capabilities.

Meta executives say Muse Spark is just the beginning, with more advanced models coming this year.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It’s free.