Meta Brings Open Models to PCs
New Muse Glimmer model runs locally as Zuckerberg pushes Washington to remove barriers facing open-weight AI developers.
Topics
News
- Temasek, Seraphim Back Pixxel in $100 Million Round
- UN Rights Chief Warns Advanced AI Could Threaten Humanity
- Tata-owned JLR to Cut About 4,000 Bobs Amid Tariff, China Pressure
- Anthropic Walks Away From $6 Billion Decart Deal
- Hyundai India Targets 20% Women in Executive Workforce by 2030
- Security Breach Drains $320 Million in Bitcoin From Liquid Network
Meta released an open-weight AI model designed to run autonomous tasks on personal computers as Chief Executive Mark Zuckerberg urged the US to reduce barriers facing domestic developers competing with China.
Muse Glimmer is a roughly 30-billion-parameter model built for local AI agents, coding and other multistep tasks. Meta released its weights under the Apache 2.0 license, allowing developers to download, modify and run the model without relying on its cloud infrastructure.
The model can process text and images, plan across several steps, call external tools and retry when commands fail. It was trained on data spanning more than 100 languages and distilled from Meta’s larger Muse Spark model, according to the company’s model card.
“Running models locally enables you to use AI anywhere, anytime, with or without an internet connection,” Meta said in its release.
Meta designed Glimmer to run on high-end consumer hardware. The full-precision version targets systems with 64 GB of memory, while quantization reduces the language model to less than 20 GB. That allows it and supporting components to fit within a 24 GB or 32 GB memory envelope, Meta said.
The release revives Meta’s open-weight push as Chinese developers including DeepSeek, Alibaba and Moonshot gain ground with downloadable models that businesses can customize and operate themselves. The most capable models from US rivals OpenAI, Anthropic and Google remain largely closed.
Zuckerberg also said Meta will release the weights for Muse Spark 1.2, its most advanced model, and has larger systems in development.
Zuckerberg used a policy statement published Monday to argue that US developers face disadvantages in areas including training-data rules and model distillation, which transfers capabilities from larger models to smaller ones.
“US policy must reduce this additional friction if we want American open source models to lead over time,” he wrote in The Future Is for Everyone. He argued that restricting access to leading foreign open-source models would reduce the quality of AI available to US users and companies without removing the hurdles facing American developers.
Meta said Muse Glimmer was evaluated under its Advanced AI Scaling Framework before being cleared for release.
“We’re releasing Muse Glimmer open weights today on Hugging Face, along with developer documentation to help you start building and running your own agents,” the company said.
The strategy gives Meta a different route from OpenAI, Anthropic and Google, whose most capable models remain largely closed. Rather than competing only on model size, Meta is betting that cheaper, customizable AI running directly on users’ devices can become another front in the race for developers.
