Meta has delivered a major blow. On April 8, 2026, the Menlo Park giant unveiled Muse Spark—the first artificial intelligence model to emerge from Meta Superintelligence Labs (MSL), the division Mark Zuckerberg created to develop "superintelligent" level AI. Muse Spark is not a simple iteration of Llama; it is, according to Meta, a "ground-up overhaul" of the company's approach to AI. The model is natively multimodal, capable of visual chain-of-thought reasoning, function calling, and multi-agent orchestration.
Muse Spark's technical capabilities are impressive. The model supports simultaneous processing of text, images, and video, with an extended context window. It integrates a tool-use system that allows it to interact with external APIs, execute code, and navigate the web autonomously. Meta claims superior performance to Claude Sonnet and comparable results to GPT-5 on reasoning and coding benchmarks, although independent evaluators note that Google Gemini 3.1 Pro still holds an advantage on certain complex reasoning tests.
Meta's strategy is clear: after dominating open-source AI with the Llama family, the company is shifting to a hybrid approach. Muse Spark will be offered both as open-source (with certain restrictions) and in a premium commercial version via Meta AI infrastructure. "We learned from Llama that openness builds the ecosystem, but we also need a revenue model to fund superintelligence R&D," said Yann LeCun, Meta's chief AI scientist, during the presentation.
The launch of Muse Spark comes amid fierce competition. OpenAI has just secured $122 billion in funding, Anthropic has unveiled Claude Mythos, and Google has launched Gemma 4 and Gemini 3.1 Pro. Meta, which invested over $65 billion in AI infrastructure in 2025 and 2026, is betting its credibility on this model. The name "Muse" symbolizes a departure from the Llama lineage and the beginning of a new era—that of "personal superintelligence," according to Meta AI's official blog.
Analysts are proceeding with caution. While Muse Spark represents undeniable progress over Llama 4, its positioning within the ecosystem remains to be clarified. Decrypt notes that Google's Gemini 3.1 Pro "still leads the pack" on several key benchmarks. The real question is adoption: will developers who massively adopted Llama switch to Muse, or will they prefer the stability of the existing ecosystem? Meta is banking on backward compatibility and a gradual migration to win them over.
