Meta released Muse Glimmer, a 30B open-weight parameter model designed for local, always-on agent workflows, alongside a follow-up essay from Mark Zuckerberg reaffirming Meta's commitment to personal superintelligence (PSI). Glimmer is optimized to run on a single RTX 3090 GPU, delivering competitive performance on key agentic benchmarks for its size class. Spark (broader-capability variant) is expected soon. The timing marks Meta's resurgence in open-weights frontier models after a period of more cautious releases following the Dreamer acquisition and Muse Code rollouts.
Zuckerberg's August 10 essay frames Meta's AI strategy around distributing superintelligent capabilities to individuals rather than centralizing power in large institutions. Core predictions include: everyone gets a personal agent understanding their goals, creation tools for expression, business/entrepreneurship tooling, personalized tutoring, and free/affordable access. He argues personal AI systems will constrain any single superintelligence from achieving monopolistic control, since distributed agents with different values check and compete against each other like a natural economy.
Strategically, Zuckerberg emphasized American infrastructure advantages (silicon design) and disadvantages (energy build speed relative to China's 1GW+/fortnight nuclear capacity), arguing export controls on chips are strategically sound but US policy should avoid delaying model releases even by months, as competitive risk from foreign labs is acute. He also proposed frontier labs share intermediate training checkpoints with government for hardening critical systems rather than waiting until training completion.
For architects, Glimmer's release on consumer GPUs signals Meta is betting local, user-controlled agents are more viable than the API-centric frontier-lab model. Longer term, Zuckerberg's emphasis on decentralized superintelligence and competitive agent ecosystems is a direct counter to fears of singleton AI—his argument that open-weights models distributed to individuals are a governance solution. The political-economic frame is new for Meta and may shape how they approach scale-ups and partnerships.