Meta’s New AI Model Prioritizes Speed Over Scale
Muse Spark, originally code-named Avocado, is intentionally designed to be small and fast rather than massive. The model can reason through complex questions in science, math, and health while requiring an order of magnitude less compute than Meta’s older midsize Llama 4 variant. Meta announced the model Wednesday, with the company’s stock climbing nearly 9% following the news.
Wang joined Meta in June as part of the company’s $14.3 billion investment in Scale AI, where he served as CEO. Meta Superintelligence Labs has rebuilt the company’s AI stack from scratch over the past nine months, according to the company’s Wednesday blog post.
Why Meta Needs This Win Now
The stakes are enormous. OpenAI and Anthropic are collectively valued at over $1 trillion, while Google’s Gemini has gained significant consumer traction. The global generative AI market is projected to grow more than 40% annually, climbing from approximately $22 billion in 2025 to nearly $325 billion by 2033, according to Grand View Research.
Meta’s previous open-source AI models failed to gain developer momentum, forcing Mark Zuckerberg to shift strategy. This time, Muse Spark launches as proprietary, though Meta says it hopes to open-source future versions.
Muse Spark Brings Multiple Operating Modes
The model powers Meta’s digital assistant across the Meta AI app and desktop website, with rollout planned for Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban Meta AI glasses in the coming weeks. Users can toggle between different modes based on query complexity. A Quick mode handles simple questions, while a Contemplating mode uses AI agents reasoning in parallel to compete with frontier models like Gemini Deep Think and GPT Pro.
Meta is also introducing a Shopping mode within the revamped AI assistant, designed to help users purchase clothing or decorate spaces by drawing from creator inspiration already circulating across Meta’s platforms.
Building a Revenue Stream from AI
Meta is experimenting with a new business model by offering third-party developers access to Muse Spark through an API. Currently, only select partners can access a private API preview, but the company plans to eventually offer paid access to a broader audience. This marks Meta’s first significant attempt to monetize AI model technology directly, diverging from its previous open-source Llama strategy.
The company is simultaneously ramping up infrastructure spending. Meta’s AI-related capital expenditures for 2026 are projected between $115 billion and $135 billion, nearly double last year’s capex, reflecting the massive computational demands of competing in frontier AI development.
What Comes Next for Meta’s AI Ambitions
Muse Spark represents Meta’s bet on efficiency over raw capability. The model will eventually power the company’s Vibes AI video feature, currently using third-party technology from Black Forest Labs. Success depends on whether developers and users embrace Meta’s efficiency-focused approach while competitors continue scaling larger models. The company acknowledged performance gaps remain in long-horizon agentic systems and coding workflows, areas it plans to address in coming iterations.
Follow Hashlytics on Bluesky, LinkedIn , Telegram and X to Get Instant Updates
