Meta bets on open models in a bid to reboot its AI strategy

Meta has announced an intention to focus on open-weight large language models, unveiling the open model Muse Glimmer and promising to open the weights of its more powerful Muse Spark 1.2 in coming weeks. CEO Mark Zuckerberg published a 6,000-word essay outlining the company's AI philosophy, aiming to differentiate Meta from OpenAI and Anthropic, which develop proprietary models and have lobbied the U.S. government against large-scale distillation and open-weight models from Chinese labs. Muse Glimmer is a 30 billion parameter model with a 128,000-token context window, distilled from Muse Spark, and is designed to run locally on user machines under the Apache 2.0 license. Muse Spark, launched in April as a closed frontier model, marked a shift from Meta's prior open approach; its 1.2 version, released August 5, included Muse Code, a coding agent that competes on cost rather than top capability. Muse Glimmer reflects a trend toward local inference to reduce reliance on big-lab models like those from Anthropic and OpenAI.