Scott Bessent Supports Meta's Muse Glimmer as Key to Innovation and Security
US Treasury Secretary Scott Bessent has publicly endorsed Meta’s release of Muse Glimmer, a new open-source AI model, describing it as a significant achievement for American innovation and a safeguard against foreign intellectual property theft. This endorsement merges two often disparate topics: the philosophy of open-source software and national security strategy.
Launched on August 10, 2026, Muse Glimmer is a 30-billion-parameter dense multimodal model that can operate on consumer-grade GPUs. It is distributed under the Apache 2.0 license, allowing users to utilize, modify, and share it freely.
The model is tailored for what the AI community refers to as "agentic workloads," which involve tasks that can run autonomously for extended periods, making decisions and recovering from failures without continuous human oversight. Muse Glimmer boasts a context window exceeding 120,000 tokens, enabling it to manage substantial information during a single session. This capability positions it competitively with several leading commercial models, with the added advantage of running on existing consumer hardware.
Meta has ensured that Muse Glimmer is compatible with consumer GPUs from NVIDIA, AMD, and Apple, contrasting with many advanced AI models that necessitate expensive data center infrastructure. This model is a distilled version of Meta’s larger, proprietary Muse Spark 1.2, sacrificing some raw performance for the ability to function on desktop machines.
Bessent’s endorsement extends beyond mere support; he explicitly linked the development of open-source AI by American firms to the broader objective of safeguarding US intellectual property from foreign threats. The geopolitical context is crucial, as tensions between the US and China regarding AI development and intellectual property have escalated, with export controls on advanced chips and scrutiny of Chinese AI labs accused of using proprietary Western datasets.
Meta has been cultivating this strategy for years through its Llama family of large language models, establishing itself as a leading corporate supporter of open-weight AI. Muse Glimmer marks an expansion into multimodal capabilities, allowing the processing of text, images, audio, and potentially other data types.
The shift to running AI locally, rather than through cloud APIs, alters the economic and power dynamics of the technology. When a model operates on personal hardware, user data remains secure, eliminating concerns over API costs and vendor control.
While the hardware requirements for Muse Glimmer are significant, necessitating a GPU with at least 24GB of VRAM for optimal performance, this still falls within the range of enthusiast-grade hardware. The Apache 2.0 license is a strategic choice, offering one of the most permissive frameworks for commercial use, unlike some of Meta’s earlier Llama releases that imposed restrictions based on company revenue.
Meta’s approach of releasing a distilled version of its larger model rather than fully open-sourcing Muse Spark 1.2 follows a common industry pattern, where companies retain their most advanced systems while providing smaller, efficient variants to showcase technical leadership and foster developer loyalty.
The implications of this competitive landscape extend beyond Meta. OpenAI has leaned towards more closed development, while Google has released some open models but keeps its flagship systems proprietary. Bessent’s public endorsement of the open-source approach adds a political dimension to what has traditionally been a business decision.
