Meta has introduced its new AI model, Muse Spark Open. The company positions the model as the most capable open-weights AI system it has developed to date.
Open-weights models offer developers different advantages over closed models in terms of cost, transparency, and customization flexibility. Meta's new model is specifically designed for complex coding processes and automation-based agent workflows.
The technical specifications of Muse Spark Open include 816 billion parameters and a 1 million token context window. This capacity aims to enable the model to handle long texts, large codebases, and multi-step tasks within the same context.
One of the notable features is that the model is designed to run locally on Apple computers. However, there is a high hardware requirement for this usage: it is stated that at least 1 TB of RAM is required to run the model locally. It is also noted that the most efficient experience will be provided on Mac Studio devices with M5 Ultra chips.
CODING AND VIDEO EDITING TESTS
Meta highlighted Muse Spark Open not only with its technical capacity but also with application examples. It was announced that the model has been integrated into various agent workflows along with the company's AI coding terminal, Muse Code.
According to the information shared, Muse Spark Open was tested on real-world coding tasks such as debugging, adding new features, and configuration on a Mastodon client. It was also reported that the model is capable of performing video editing operations such as reviewing and cutting footage in DaVinci Resolve.
This move by Meta is considered one of the developments that could strengthen the competition for running large-scale AI models not only on cloud services but also on high-performance local systems.
News Source: 12punto
