Meta released Muse Glimmer, a 30-billion-parameter open-weight model licensed under Apache 2.0, designed to run AI agents locally on consumer hardware. It supports text and images, was trained in more than 100 languages, and is intended to operate offline on a single consumer GPU.
Glimmer is essentially an open version of Meta's most powerful closed model, Muse Spark, which debuted in April. The model can call tools, write and debug code, work with files and screenshots, and run multi-step tasks locally, reflecting CEO Mark Zuckerberg's vision of 'personal superintelligence.'
By running agents on a user's device instead of the cloud, Glimmer offers a more privacy-sensitive personal-assistant approach and opens powerful agentic AI to individual developers, potentially shifting dynamics away from cloud-dominated AI providers.

Meta released Muse Glimmer, a 30-billion-parameter open-weight model licensed under Apache 2.0, designed to run AI agents locally on consumer hardware. It supports text and images, was trained in more than 100 languages, and is intended to operate offline on a single consumer GPU.

Glimmer is essentially an open version of Meta's most powerful closed model, Muse Spark, which debuted in April. The model can call tools, write and debug code, work with files and screenshots, and run multi-step tasks locally, reflecting CEO Mark Zuckerberg's vision of 'personal superintelligence.'

By running agents on a user's device instead of the cloud, Glimmer offers a more privacy-sensitive personal-assistant approach and opens powerful agentic AI to individual developers, potentially shifting dynamics away from cloud-dominated AI providers.

📰 Source: TechCrunch
techcrunch.com ↗
Was this article useful?