Meta Platforms Inc. announced Monday the open-sourcing of its Muse Spark 1.2, its most advanced AI model to date, alongside the release of Muse Glimmer, a smaller open-weight model designed to operate on consumer devices such as laptops. The company’s shares rose 2.1% in premarket trading following the announcement.
Meta CEO Mark Zuckerberg framed the move as part of a broader push for open-weight AI, where model parameters are publicly accessible for customization and local deployment. In a 6,500-word essay published Monday, Zuckerberg criticized the concentration of AI power among closed-model developers, including rivals like OpenAI and Anthropic, while highlighting the competitive advantages of open-source approaches. The models are developed under Meta’s Superintelligence Labs, established in 2024, with Zuckerberg asserting that the company’s $145 billion capital expenditure forecast for 2025 would yield long-term returns.
Muse Glimmer is described as an agentic model, capable of performing tasks autonomously, and was trained using a distillation process from the larger Muse Spark 1.2. The smaller model is optimized for efficiency, enabling operation on standard computing hardware without cloud dependency. Meta also announced plans to release an open-weight version of Muse Spark 1.2 in the near future, expanding access to its most powerful model.
The announcement follows a trend among some AI developers, particularly in China, where companies like Alibaba, DeepSeek, and Moonshot have released competitive open-weight models. Zuckerberg’s essay emphasized the need for U.S. policy support to counter foreign advantages, citing restrictions on training data that U.S. labs face compared to international competitors. The company did not disclose specific performance benchmarks for the new models but stated they were designed to rival leading closed systems.
Meta’s stock, which has declined approximately 10% year-to-date, saw a modest premarket gain following the announcement. Investors continue to scrutinize the company’s spending strategy amid concerns over its ability to compete with major AI labs. Zuckerberg reiterated that the Superintelligence Labs’ progress would justify the substantial capital outlay, though no timeline for profitability was provided.
The open-sourcing initiative aligns with Meta’s broader strategy to position itself as a leader in accessible AI development, contrasting with the proprietary models dominating much of the industry.