Meta has introduced Muse Spark Open, a model with open weights. The company calls it the most capable open-weight model in the US. It is multimodal and built for heavy coding and agent-style work.
What we know about the model
It has 816 billion parameters and a context window of up to one million tokens. The idea is to run it locally on Apple hardware, ideally a Mac Studio with the M5 Ultra chip. There is a catch: according to Engadget, which TechCrunch cites, it needs at least 1 TB of RAM. That is not a laptop scenario.
How it was shown
Meta paired the model with its coding terminal, Muse Code, and used it to work on a Mastodon client: fixing bugs, adding features and producing a full build. In a second demo, the model worked with DaVinci Resolve to review and crop video footage. API access is already open on several platforms.
What is still missing
Meta says the model beats other popular open-weight models, but the announcement gives no specific scores. The license, API pricing and Apple's exact role are not spelled out either. For now it is more of a claim than a verified result.