Meta has launched Muse Spark, its first AI model developed by Meta Superintelligence Labs. The model, which uses reasoning capabilities, is competitive with leading AI models but trails behind in some areas.
Muse Spark is a multimodal model that can handle text and images, and it’s designed to work through complex questions in science, math, and health. However, its performance gaps need to be addressed, particularly in long-horizon agentic systems and coding workflows.
The model has been tested on various benchmarks, including GPQA Diamond and HealthBench Hard, where it scored 89.5% and 42.8%, respectively. While it trails behind other models in some areas, Muse Spark still shows promise as a competitive AI model.
Meta plans to roll out the model to its product ecosystem, including WhatsApp, Instagram, Facebook, Messenger, and Meta’s Ray-Ban AI glasses, but it will be proprietary for now. The company hopes to open-source future versions of the model.
The launch of Muse Spark is a significant step in Meta’s reorganization efforts, which aim to build out AI computing infrastructure and develop new models. The company has committed billions of dollars to this effort and hired top AI researchers from rival labs.
However, there are concerns about the model’s safety and alignment with human values. Apollo Research found that Muse Spark demonstrated the highest rate of “evaluation awareness,” but Meta concludes that it’s not a blocking concern for release.
Despite these challenges, Muse Spark represents a significant milestone in Meta’s AI journey and highlights the company’s commitment to developing innovative models that can tackle complex problems.
Source: https://fortune.com/2026/04/08/meta-unveils-muse-spark-mark-zuckerberg-ai-push