Meta has revealed Muse Spark, the first model from its new division Meta SuperIntelligence Labs, marking its return to the race for artificial intelligence (AI) after a significant investment of $14.3 billion in hiring Alexandr Wang, founder of Scale AI. This launch is significant, as Muse Spark is the first foundational model since Llama 4 in April 2025, and represents a considerable effort by the company to rebuild its AI strategy.
Muse Spark is here
According to Meta, Muse Spark is notably more efficient than its predecessor, Llama 4 Maverick, using ten times less computing power. Through various benchmarks, Muse Spark is competitively positioned in the realm of multimodal reasoning, surpassing rivals like Claude Opus 4.6 and OpenAI GPT-5.4. However, it shows weaknesses in abstract thinking and agentic programming, critical areas in the current development of AI.
An innovative aspect of Muse Spark is its “contemplative mode,” which allows for the collaboration of multiple agents reasoning in parallel, improving latency and the quality of responses in complex tasks. This feature promises to reduce wait times for users, which is essential in the everyday use of AI technology.
Despite expectations, there are concerns about whether Muse Spark will be able to meet the high market expectations, given that its overall performance still does not surpass that of its closest competitors. Additionally, the model breaks with Meta’s open-source tradition, being proprietary and closed, which could limit its adoption and future evaluations.
The initial results have been encouraging, as Meta’s shares rose by 6.5% after the announcement. However, it is expected that Muse Spark will demonstrate its true value in an arena where competition is fierce and users are increasingly demanding.
