Meta resets its AI race

Meta Platforms has unveiled Muse Spark, its first new flagship artificial intelligence model in about a year, marking a pivotal moment in Mark Zuckerberg’s attempt to restore the company’s standing against OpenAI, Google and Anthropic after a period of criticism around its earlier Llama releases. Muse Spark began rolling out on 8 April through the Meta AI app and the meta. ai website in the United States, with integration into WhatsApp, Instagram, Facebook, Messenger and Meta’s AI glasses due in the coming weeks.

The launch is more than a product update. It is the first public output from Meta Superintelligence Labs, the reorganised unit built after Zuckerberg embarked on a costly reset of the company’s AI effort. Reuters and other outlets reported that Meta’s push included a $14.3 billion investment tied to Scale AI and the recruitment of its chief executive, Alexandr Wang, as Meta sought to close the gap with rivals whose models had moved ahead in reasoning, coding and enterprise adoption.

Meta described Muse Spark as a natively multimodal reasoning model built for consumer use rather than as a purely developer-facing system. The company says it can process text and images, use tools, and employ multiple specialised agents on harder tasks. On its own materials, Meta positions the model around practical assistance, from shopping and discovery to health-related queries and everyday decision-making, part of Zuckerberg’s wider push towards what he calls “personal superintelligence”.

That ambition comes with an unusual level of secrecy for a company that had spent years championing open AI models. Meta did not disclose Muse Spark’s size, a metric commonly used to gauge a model’s computing scale and compare it with rivals. It has also kept access tighter than in the Llama era, offering only a limited preview to partners through an API while keeping the model itself closed. That decision signals a shift in strategy as Meta seeks tighter product control and faster consumer deployment.

Early evidence suggests the model is competitive, though not dominant. Reuters reported that Muse Spark ranked fourth on the Artificial Analysis intelligence index, placing it near the frontier but behind the very top systems. Meta’s own evaluation material says the model performs strongly in multimodal perception, reasoning, health and agentic tasks, while external reporting indicates that coding and some abstract reasoning remain weaker areas. That mix is important because it suggests Meta has built a model strong enough for mainstream consumer products without yet matching the leading edge across every category.

The commercial logic is clear. Unlike enterprise-focused rivals that earn revenue from developer platforms and subscriptions, Meta is trying to weave AI into an advertising and commerce ecosystem already used by billions of people. Reporting from Reuters and other publications indicates the company sees Muse Spark as a future engine for product discovery, shopping assistance, visual recommendations and task completion across its apps and devices. If that strategy works, Meta may not need to win every benchmark contest to build a profitable AI business.

Investors have so far treated the launch as a relief rather than a final verdict. Market commentary from several outlets showed upbeat responses from Wall Street analysts who viewed the release as proof that Meta’s heavy spending was beginning to yield tangible products. The more cautious view, however, is that enthusiasm around launch week does not settle the harder question of whether users will change habits, trust Meta with more sensitive tasks, or prefer its assistant over entrenched competitors.

There are also policy and credibility questions. Axios noted that privacy terms around Meta AI remain a point of concern because the company may use information shared with the system to improve products and services. For a model that Meta wants to position closer to personal decision-making, that issue could become more politically and commercially significant, especially in health, shopping and communication settings where user expectations are higher.