Meta, which has been attracting Silicon Valley AI talent with the extraordinary offer of a maximum annual salary of USD 100 million (about KRW 1480 billion) in a bid to develop “superintelligence” surpassing humans, has finally announced its first result.

On the 8th (local time), Meta stated that its Meta Superintelligence Lab (MSL), led by Alexander Wang, founder of Scale AI and Meta’s Chief AI Officer (CAIO), unveiled its first AI model, “Muse Spark.” The industry views this model as putting Meta back into the “AI race” among Big Tech, after it had fallen far behind with its earlier AI model “LLaMA,” which drew heavy criticism. On the day of the announcement, Meta’s share price closed at USD 612.42, up 6.5% from the previous day.

● Era of ‘LLaMA’ ends, era of ‘Muse’ begins

“Muse Spark” has drawn attention as the first test case from MSL, the core of Meta AI. Mark Zuckerberg, Meta’s Chief Executive Officer (CEO), invested USD 14.3 billion (about KRW 20 trillion) last June in Scale AI, a company specializing in AI data labeling (refining, classifying, and annotating training data for AI), and recruited its founder Wang, known as a genius developer, as Meta’s CAIO. Having finally revealed the outcome, CAIO Wang wrote on social networking service (SNS) X on the 8th, “Nine months ago, Superintelligence Labs rebuilt the AI stack from scratch,” adding, “Muse Spark is the result of those efforts, and it will now lead Meta AI.”

According to performance indicators disclosed by Meta, Muse Spark is on par with, or even surpasses in some metrics, existing top-tier models such as OpenAI’s “GPT-5.4,” Google’s “Gemini 3.1 Pro,” and Anthropic’s “Claude Opus 4.6.” According to Bloomberg, a Meta executive stated that while the model provides excellent answers to questions related to science, health, and mathematics, its coding capabilities are relatively weak.

A “Deliberation Mode,” in which multiple AI agents collaborate to derive an optimal answer, has also been integrated. Deliberation Mode scored 50.2% on the “Humanity’s Last Exam (HLE)” benchmark, which measures expert-level reasoning ability, exceeding Google’s Gemini 3.1 Deepsync (48.4%). Meta announced that Muse Spark will be used in “WhatsApp,” “Instagram,” “Facebook,” and Meta’s “AI glasses” in the future. Unlike LLaMA, which was released as open source, the new model will be operated as a closed system for the time being.

● Is a full-scale AGI development race beginning?

Industry observers note that although the new model falls far short of the superintelligence Meta ultimately aims to develop, it is significant in having raised performance to the level of major AI models within a short period. Meta, which had been considered completely outdated and “past its prime,” has made a spectacular return with the Muse series. CEO Zuckerberg said, “This model is important in that it shows how quickly we are progressing. We will continue to release new models and push the boundaries of the technology throughout the year,” signaling that follow-up models will be launched.

While current AI excels in specific domains, artificial general intelligence (AGI) refers to a stage where it can think in a human-like way across all fields, and artificial superintelligence (ASI) denotes a level that surpasses humans. As Meta ultimately targets ASI development, the market is seeing a growing trend toward the materialization of AGI competition. Ray Kurzweil, author of “The Singularity Is Near,” recently predicted at an AI conference that “AGI will certainly be realized by 2029 at the latest.”

Popular News