Meta unveils Muse Spark model in major AI overhaul

Meta announced on Wednesday the release of a new AI model called Muse Spark, describing it as the "first step" in a broader overhaul of its artificial intelligence strategy.

Muse Spark is the first model developed by Meta Superintelligence Labs, a division established last year after CEO Mark Zuckerberg reportedly expressed dissatisfaction with the pace of progress on Meta's AI initiatives, including its Llama models, which were seen as trailing competitors like OpenAI's ChatGPT and Anthropic's Claude.

To strengthen its AI efforts, Meta brought in Alexandr Wang, the former co-founder and CEO of Scale AI, to lead the new lab. The company also invested $14.3 billion in Scale AI, acquiring a 49% stake as part of its strategy to enhance data capabilities and model development.

With Muse Spark now available on the web and in the Meta AI app, the company aims to demonstrate the potential of its restructured AI team. Meta indicated that the model will continue to evolve, with plans to introduce a feature called "Contemplating" mode, designed to handle more complex tasks.

This mode will rely on multiple AI agents working simultaneously on the same problem, enabling the system to deliver faster, more refined results. According to Meta, this approach enables increased reasoning capacity without significantly increasing response time.

"To spend more test-time reasoning without drastically increasing latency, we can scale the number of parallel agents that collaborate to solve hard problems," the company explained in its announcement.

While competitors often restrict advanced AI capabilities behind paid subscriptions, Meta has not yet clarified whether Muse Spark will adopt a similar monetisation model.

The company is also aligning with a growing industry focus on health-related applications. Meta noted that Muse Spark could assist users with health-related queries, an area where other AI developers are also expanding their offerings.

However, this direction may raise privacy concerns. Access to Muse Spark requires users to log in with an existing Meta account, such as Facebook or Instagram. Although Meta has not explicitly stated that the AI will use personal account data, the company's history of utilising public user data for training suggests such concerns may arise, particularly given that Muse Spark is positioned as a personalised AI system.

Meta also highlighted the model's strength in handling visual STEM-related queries, which could enable interactive use cases such as creating simple games or assisting with troubleshooting household devices.

In addition to its investment in Scale AI and the appointment of Wang, Meta has recruited researchers from leading AI organisations, including OpenAI, Anthropic, and Google, signalling a significant push to compete at the highest level in the AI space.

Zuckerberg emphasised the company's long-term vision, stating that future models will continue to advance the boundaries of AI capabilities, including the release of new open-source systems. He added that Meta is focused on building AI products that go beyond answering questions and can actively perform tasks on users' behalf. The launch of Muse Spark represents a critical moment for Meta as it seeks to establish itself as a stronger competitor in the rapidly evolving AI industry.