Meta has launched a new 30-billion parameter AI model named Muse Glimmer, designed to run on consumer devices. The model focuses on tasks such as AI agents, coding, and function calling.
Taking a significant step in the AI landscape, Meta has launched a new open-weight AI model named Muse Glimmer. The company aims to create AI models that can run directly on everyday consumer computers and devices, rather than relying on massive data centers or cloud servers. Meta’s new model is specifically designed for tasks such as local AI agents, coding, and function calling.
A 30-Billion Parameter AI Model
Muse Glimmer features approximately 30 billion parameters. Despite this scale, it has been optimized to run on standard consumer-grade Macs or PCs equipped with typical consumer GPUs. While large AI models usually require multiple powerful GPUs and massive data centers, Meta’s model seeks to make AI processing on local devices a practical reality.
This means developers will be able to build AI agents and applications that operate directly on the user’s computer, eliminating the need to send data to cloud servers for every minor task.
Special Focus on AI Agents and Coding
Muse Glimmer is tailored for use cases like AI agents, coding, and function calling. AI agents are systems capable of executing multi-step actions to complete a task, rather than simply answering questions.
For instance, a local AI agent could eventually organize files on a computer, write code, perform actions within software, or utilize other tools. However, actual performance will depend on the device’s hardware and the optimizations implemented by developers.
Available Under the Apache 2.0 License
Meta has released the model weights for Muse Glimmer under the Apache 2.0 license. Developers can download the model, integrate it into their applications, modify it as needed, and run it on their own hardware. The model is available via Hugging Face.
This move is part of Meta’s broader strategy to make AI models more openly accessible to developers. Meta’s approach differs from cloud-based AI
Currently, for popular AI systems like ChatGPT, a user’s query or command travels via the internet to the company’s data center. Processing takes place there on powerful GPUs, and the response is then sent back to the user’s device.
Local models like Muse Glimmer could transform this process. Running an AI model directly on the device can reduce reliance on cloud servers for every task. In certain situations, this can lead to faster response times and provide users with greater control over their data.
However, the benefits of local AI will largely depend on the device’s computing capabilities, its GPU, and the optimization of the model itself.
Meta also announces Muse Spark 1.2
Alongside Muse Glimmer, Meta CEO Mark Zuckerberg has also announced Muse Spark 1.2. The company plans to make its model weights publicly available as well. Muse Glimmer is seen as a step towards creating smaller, more practical AI models designed to run on consumer hardware.