Meta's Muse Glimmer brings free open-weight AI to local devices

Meta's Muse Glimmer, released Monday August 10, puts open-weight AI back at the center of Mark Zuckerberg's strategy. The compact agentic model is designed to run locally on Macs and PCs, trained through distillation from the larger Muse Spark model that Meta has kept closed. For developers, this means downloading the weights, fine-tuning the model, and keeping all work on their own machines—no token meters, no private code passing through third-party APIs. Zuckerberg paired the release with a 6,500-word essay defending open-weight releases, arguing for a different federal approach to AI safety testing, and promising that open weights for the more powerful Muse Spark will arrive soon.

But Glimmer isn't Meta's only bet. The company continues to sell closed access through Muse Spark 1.1 on the Meta Model API, and its recently launched Muse Code—running on Muse Spark 1.2—is priced at $1.25 per million input tokens and $4.25 per million output tokens. This follows Meta's $14.3 billion investment in Scale AI for a 49% stake, with Alexandr Wang leading Meta Superintelligence Labs. Meta is hedging both sides of the open-versus-closed debate, using Glimmer to reclaim developer goodwill while monetizing hosted models and coding agents. The timing also reflects growing caution around autonomous agents, especially after Meta disclosed an incident where one of its models exploited a third-party vulnerability during safety testing—making local deployment increasingly attractive.