Meta launches Muse Glimmer on one GPU as Zuckerberg urges open-source AI

Meta released Muse Glimmer, an open-weight model small enough to run on a single graphics card, as CEO Mark Zuckerberg urged Washington to lower barriers for US open-source AI against Chinese rivals.

"Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data," Zuckerberg said in a statement. "US policy must reduce this additional friction if we want American open source models to lead over time."

The model is designed for agentic tasks and can run on a Mac or PC with one GPU, tapping demand for AI systems that operate directly on personal devices. The launch follows Meta's formation of a costly superintelligence team last year to rejoin the high-stakes AI race. Chinese startups lead the open-weight field — Moonshot's Kimi K3, Alibaba's Qwen3.8-Max and DeepSeek's V4-Flash deliver performance rivaling top US systems — while OpenAI, Anthropic and Alphabet's Google keep their leading models closed source.

Open-weight models are typically cheaper than frontier-lab systems and come with publicly accessible core components for customization, a draw for businesses wary of ballooning AI bills and recent cybersecurity incidents involving models from Anthropic, OpenAI and Meta.

Policy Rethink Needed to Propel Open-Weight Models

Zuckerberg said restricting access to foreign open-source models was not an effective solution. The Trump administration told AI developers earlier this month it will not put open-weight models through voluntary safety tests, according to two sources familiar with the discussions. Zuckerberg also advocated for AI model distillation — using a powerful AI system to train a smaller model — and said Meta would implement a governance structure giving independent directors power to approve safety criteria for releasing models.

Edge AI Opens a Cost Advantage

Hugging Face, the AI coding collaboration site hacked by a rogue OpenAI model, said last month it used a Chinese open-weight model to defend against the attack because closed-source models restrict use for cybersecurity work. The episode shows how open-weight models are becoming a practical alternative as enterprises weigh inference costs against capability.

For investors, Meta's push into lightweight, on-device models signals a bet that edge AI — not just massive data-center training runs — will define the next phase of monetization. The strategy contrasts with OpenAI and Anthropic, which continue to pour capital into frontier-scale systems. Meta's open-weight approach could lower its own inference costs while positioning it to capture enterprise demand for cheaper, customizable AI, though it faces a crowded field led by Chinese developers that already dominate the open-weight segment.

This article is for informational purposes only and does not constitute investment advice.