Meta Launches Muse Spark AI Model from Superintelligence Labs

Muse Spark is designed as a small and fast model capable of reasoning through complex questions in science, math, and health. Meta stated that improved training techniques and rebuilt infrastructure enabled creation of models as capable as previous midsize variants but with significantly less compute.

The company positions it for competitive performance in multimodal perception, reasoning, health, and agentic tasks.

Labs was established following the hiring of Alexandr Wang in June, who previously served as CEO of Scale AI.

3 billion investment in Scale AI nine months ago. The labs rebuilt Meta's AI stack over the last nine months, accelerating development cycles. The release follows the disappointing performance of Meta's previous open-source models in April, which failed to attract developers.

CEO Mark Zuckerberg adjusted the strategy in response. Muse Spark marks a shift to a proprietary model, though Meta expressed intent to open-source future versions.

now powers Meta's digital assistant across its platforms.

It will integrate into Facebook, Instagram, WhatsApp, and Messenger within weeks, as well as Ray-Ban Meta AI glasses. The model will also support the Vibes AI video feature in the Meta AI app, replacing third-party models. Users can select modes based on query complexity: Instant for simple questions, Thinking for tasks like analyzing legal documents or nutritional information from photos, and Contemplating for advanced reasoning using multiple AI agents.

A new Shopping mode assists with clothing purchases and room decoration by drawing from content in Meta's apps. Meta offers access to Muse Spark via a private API preview for select partners, with plans for paid API access to wider audiences later. The model lags rivals in coding ability, according to benchmarks.

Meta continues investments in areas like long-horizon agentic systems and coding workflows.

aims to compete with OpenAI, Anthropic, and Google in the AI market, where OpenAI and Anthropic hold combined valuations over $1 trillion.

Google's Gemini has gained consumer traction. The global generative AI market is projected to grow from $22 billion in 2025 to $325 billion by 2033. Meta's AI-related capital expenditures for 2026 are forecasted at $115 billion to $135 billion, nearly double the previous year's total.

The company has used AI to enhance advertising and operational efficiencies but has not yet achieved significant market share in AI models. Meta shares rose 8% during midday trading on Wednesday.