Treasury Secretary Scott Bessent Welcomes Meta's Release of Muse Glimmer Open Source Model
U.S. Treasury Secretary Scott Bessent posted that he welcomes Meta's release of Muse Glimmer, marking another victory for American innovation. Maintaining U.S. leadership in AI means advancing both open-source and closed-source models, ensuring a trustworthy foundation for the future.
Muse Glimmer is a 30 billion parameter open-source agent model released by Meta, with weights available under the Apache 2.0 license, optimized for local operation on consumer-grade GPUs in Macs or PCs.
The model is designed for always-on local agent workflows, supporting scenarios such as function calls, local programming, and evaluation, and offers integration with tools like llama.cpp and MLX.
Meta positions it as distilled from larger models, emphasizing the privacy and low-cost advantages of local deployment.
In terms of market mechanisms, government leaders publicly support open-source AI models, reinforcing the U.S. strategy on both open and closed tracks, with funding and developers leaning towards locally runnable open-source weights, putting pressure on purely closed cloud dependency paths.
Bessent emphasized the importance of advancing both to maintain leadership.
ABAB AI Insight
Scott Bessent, as Treasury Secretary, directly endorses Meta's open-source model, viewing open-source weights as a component of U.S. AI leadership rather than an opposing force.
The capital path reflects the government's public statements reducing policy uncertainty around open-source models, motivated by encouraging companies to invest in both open and closed routes, avoiding single-path lock-in.
Similar to the previous balanced stance of the U.S. government towards open-source AI, it is currently at a stage where open-source local models become key carriers for device-side AI.
Essentially, this belongs to the diversification of technological routes, with the mechanism being that as superintelligent competition intensifies, the parallel development of open-source and closed-source can maximize innovation speed and controllability.
ABAB News · Cognitive Laws
- Open-source and closed-source must advance simultaneously
- Locally runnable models are the answer for privacy and cost
- Government endorsement can accelerate the adoption of open-source weights.