Sri Lanka
USA
Canada
India
UK
UAE
Meta Platforms introduced Muse Spark on Wednesday, marking the debut of the artificial intelligence model developed by the specialized superintelligence team it assembled last year to close the gap with competitors in the AI landscape. Following the announcement, Meta’s shares rose nearly 7%, reflecting investor optimism amid growing pressure on U.S. tech giants to show returns on their sizable AI investments.
The stakes for Meta are particularly high after hiring Scale AI CEO Alex Wang last year in a $14.3 billion deal and offering select engineers compensation packages worth hundreds of millions to establish a superintelligence division. This effort aims to elevate Meta’s position in AI, especially after the underwhelming reception of its Llama 4 models early last year. Superintelligence refers to AI systems capable of outperforming human reasoning. Muse Spark is the first in a series of internal models, codenamed Avocado, developed by this team.
Initially, Muse Spark will only be accessible through the Meta AI app and website. Over the coming weeks, it is expected to replace the existing Llama models that power chatbots on WhatsApp, Instagram, Facebook, and Meta’s smart glasses. Meta has not disclosed the model’s size, an important metric for comparing computing power across AI systems. Unlike prior Llama releases, Muse Spark is being shared only as a private preview with select partners.
The company described Muse Spark as “small and fast by design,” yet capable of handling complex queries in science, mathematics, and healthcare. Independent evaluations suggest that while Muse Spark matches top-tier AI models from Google, OpenAI, and Anthropic in areas such as language and visual comprehension, it trails in coding and abstract reasoning. The model ranked fourth on a comprehensive AI performance index compiled by Artificial Analysis.
Meta CEO Mark Zuckerberg had set modest expectations for the model, noting that early versions would primarily demonstrate the rapid progress of the team. Wang, overseeing the superintelligence division, acknowledged that “rough edges” remain but emphasized that larger, more capable versions of the model are in development, some of which may be publicly released.
The release also highlights Meta’s strategy for monetizing AI. Upcoming features in Muse Spark’s chatbot include shopping tools that guide users directly to purchasable products. Beyond commerce, Muse Spark is designed to assist with everyday tasks—such as estimating calories from food photos or visualizing objects in real spaces—potentially enhancing engagement across Meta’s 3.5 billion social media users.
Additionally, Muse Spark includes a Contemplating Mode, allowing multiple AI agents to work simultaneously to increase reasoning capabilities. This feature positions Muse Spark to rival extended thinking capabilities seen in Google’s Gemini Deep Think and OpenAI’s GPT Pro, enabling applications like collaboratively planning a family vacation with one agent drafting itineraries and another researching kid-friendly activities.
