Meta has launched its new open-weight AI model, Muse Spark Open, which it claims is the most powerful of its kind developed in the United States. Unlike closed models such as Anthropic’s Claude or OpenAI’s ChatGPT, this model allows developers to download and customize its parameters.
In recent years, Chinese companies have taken the lead in the open-weight AI market with models like Alibaba’s Qwen and Z.ai’s GLM. However, Meta asserts that Muse Spark Open is now a formidable competitor, specifically designed for coding tasks, complex software engineering, and multi-step reasoning.
Meta’s Chief of AI, Alexandr Wang, highlighted the model’s robust safeguards in a blogpost. Furthermore, Muse Spark Open is optimized to operate efficiently on Apple’s advanced M5 Ultra chips, signifying a noteworthy collaboration between Meta and Apple. Meta CEO Mark Zuckerberg noted on X that the model performs exceptionally well on new Mac Studios, indicating the tech giants’ successful partnership.
Apple’s new CEO John Ternus praised the capabilities of Muse Spark Open, emphasizing the impressive efficiency of running such a sophisticated model directly on desktop systems.
Despite its capabilities, Muse Spark Open requires considerable computing resources. Meta specified that users can effectively create a “supercomputer” setup using two Mac Studios equipped with M5 Ultra chips.
The release of Muse Spark Open follows Meta’s recent announcement of two closed models, Wonder and Wonder Full, which are positioned to compete with leading models like ChatGPT and Claude.
Initially focused on open models, Meta is now striving to match the advancements made by OpenAI and Anthropic, having shifted its strategy after the earlier, less competitive Llama model.
