SAN FRANCISCO — Meta Platforms launched Muse Glimmer, an open-weight, agentic AI model capable of running on consumer devices such as a Mac or PC. The company said Glimmer handles complex reasoning and agentic tasks using a single graphics card.

The release advances CEO Mark Zuckerberg's case for broad access to advanced AI. Zuckerberg made that argument in a 6,500-word essay titled "The Future Is for Everyone," published Monday, arguing against restricting advanced AI to a select group of individuals or companies.

Meta's Superintelligence division, led by former Scale AI CEO Alexandr Wang, developed Muse Glimmer. The model was trained using a distillation process, learning from Meta's larger Muse Spark teacher model.

Muse Spark, Meta's most powerful AI offering, debuted earlier this year as a closed model. Meta plans to release an open-weight version of Muse Spark soon.

The move positions Meta within an industry debate over open-weight AI. Major AI laboratories, including Meta, signed a public letter last month advocating for open-weight models, following U.S. government suggestions about sanctioning Chinese open models derived from U.S. counterparts.

Zuckerberg addressed that policy friction in his essay, saying foreign laboratories currently hold advantages due to fewer restrictions on training data. He wrote that U.S. policy must reduce this friction for American open-source models to lead over time.

Open-weight models allow users to download and customize AI locally rather than relying on cloud-based services, offering a cheaper alternative for developers — a strategy echoed by Chinese AI firms such as Moonshot and DeepSeek.

Meta has a history with open-source models, including its Llama series first launched in 2023. The company now releases Glimmer under a permissive license.

Muse Spark remains competitive on several large language model benchmarks, though the latest version trails overall behind models from Anthropic and OpenAI. Glimmer shows strong performance within its size class on those same benchmarks.

By enabling local execution, Meta reduces developer reliance on costly cloud infrastructure. That contrasts with proprietary, cloud-hosted AI business models that generate revenue through subscription fees or compute usage.