Meta has announced a new open-weight model specifically designed to power agentic AI on local devices, according to a tweet from Mark Zuckerberg. The model aims to give developers the ability to run autonomous AI agents directly on edge hardware, reducing reliance on cloud infrastructure for inference.

The open-weight release follows Meta’s broader strategy of releasing large language models under permissive licenses. By targeting local execution, the company hopes to enable privacy-preserving applications where user data remains on-device. The model is expected to be optimized for both performance and efficiency, making it suitable for smartphones, laptops, and IoT devices.

Early comments on Hacker News highlight excitement about the potential for truly offline AI agents, though some question whether the model can match cloud-based alternatives. The announcement places Meta in direct competition with other open-weight initiatives from companies like Microsoft and Google. The full technical details and benchmark results have not yet been published.

Zuckerberg’s tweet garnered 16 points and 2 comments on Hacker News at the time of reporting, indicating modest early interest. Developers are encouraged to watch for the model’s release on Meta’s official repositories.