Meta officially released its first self-developed high-performance AI model, Muse Spark, marking a substantive step in Zuckerberg's "Superintelligence" strategy under a $100 billion investment. As the debut of the Meta Superintelligence Labs, Muse Spark is defined as an assistant with world-class intelligence, indicating that Meta is trying to overcome its lag behind OpenAI and Anthropic in the field of foundational models.
The model was developed by Alexandr Wang, founder of Scale AI, and has already been integrated into Meta AI services. Technically, Muse Spark uses a multi-agent collaboration mechanism, enabling it to handle complex, long-chain tasks such as travel planning and destination comparisons simultaneously. It also excels in visual understanding, capable of analyzing images in real-time and providing in-depth interpretations.
Although Muse Spark is not open-source yet, Meta revealed that future versions will gradually return to an open-source approach. This strategic shift is supported by substantial financial resources: Meta expects AI-related spending to rise to between $115 billion and $135 billion in 2026, far exceeding $72.2 billion in 2025. Currently, the high R&D investment has significantly translated into financial growth, with Meta's revenue increasing by 22% in 2025.
Industry experts generally believe that Muse Spark's "deep thinking mode" is fully competing with Gemini Deep Think and GPT Pro. This release not only reshapes Meta's competitive position in the reasoning model field but also indicates that the AI competition among tech giants has shifted from mere parameter size battles to ecological breakthroughs in multimodal perception and complex reasoning capabilities.
