Meta releases Muse Spark AI model as it tries to close gap with Google, OpenAI and Anthropic

Meta released Muse Spark on Wednesday, its first AI model from the $14.3 billion superintelligence team built to compete with OpenAI and Google. Stock rose 7%, though the model lagged rivals in coding and reasoning tasks.

Published on: Apr 09, 2026
Meta releases Muse Spark AI model as it tries to close gap with Google, OpenAI and Anthropic

Meta releases first AI model from $14.3 billion superintelligence team

Meta unveiled Muse Spark on Wednesday, its first generative AI model in a year and the initial release from a costly team assembled to catch up with rivals in frontier AI development. The company's stock rose nearly 7% on the announcement.

The model comes after Meta hired Scale AI CEO Alex Wang last year under a $14.3 billion deal and offered engineers compensation packages in the hundreds of millions to staff a new superintelligence unit. The move was designed to revive Meta's standing in AI after disappointing results from its Llama 4 models early last year.

Muse Spark will initially be available only on Meta's AI app and website. Within weeks, it will replace existing Llama models powering chatbots on WhatsApp, Instagram, Facebook, and Meta's smart glasses.

Performance shows mixed results

Independent tests from evaluation firm Artificial Analysis found Muse Spark matching top models from Google, OpenAI, and Anthropic in language and visual understanding. The model tied for fourth place on a broad index of AI capability tests.

Performance lagged in coding and abstract reasoning tasks. Meta did not disclose the model's size, a standard metric for comparing AI systems' computing power.

Alex Wang, who runs the superintelligence team, acknowledged "rough edges" in the model's behavior on social media Wednesday. He said larger versions are in development and that Meta plans to release at least some openly.

Features and commercial direction

Muse Spark can help users estimate calories from photos, superimpose images to preview how items look in spaces, and reason through complex questions in science, math, and health. A "Contemplating Mode" runs multiple agents simultaneously to boost reasoning capability, similar to Google's Gemini Deep Think and OpenAI's GPT Pro.

Meta is embedding shopping features directly into its AI chatbot that direct users to purchasable products. The company is betting that applying AI to everyday tasks will increase engagement across its 3.5 billion social media users-a reach advantage over smaller competitors.

CEO Mark Zuckerberg had set modest expectations in January, telling investors the team's early models "will be good but, more importantly, will show the rapid trajectory that we're on." He said Meta expected to "steadily push the frontier over the course of the year as we continue to release new models."


Get Daily AI News

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)