Meta Platforms said it will open the weights for its latest large-language model, Muse Spark 1.2, and introduce a new family of open-source models, Muse Glimmer, designed to run on laptops—an on-device push aimed at lowering costs and improving responsiveness. The moves, detailed by CEO Mark Zuckerberg in a video and a lengthy essay, sharpen Meta’s rivalry with OpenAI and Anthropic and seek to reassure investors about heavy AI spending; Meta shares rose about 2% premarket Monday but are down roughly 10% this year. Zuckerberg pressed Washington to ease constraints on training data and clarify rules around model distillation to keep U.S. open-source efforts competitive with Chinese rivals such as Alibaba, DeepSeek and Moonshot. He also warned against concentrating AI power among a few firms, arguing for widely distributed “personal agents.” The initiatives come as Meta forecasts up to $145 billion in 2026 capital expenditures tied to AI infrastructure and looks to demonstrate traction at its Meta Superintelligence Labs.
Related articles:
– Meta introduces Llama 3 open models
– Apple unveils on-device “Apple Intelligence” for iPhone, iPad, and Mac
– Anthropic releases Claude 3.5 Sonnet
– Alibaba’s Qwen open-source large language models
– Meta’s Llama 2 open-source release and ecosystem
