AI
Meta releases Muse Spark Open, an 816B model with 1M-token memory
It reportedly runs on two Mac Studios, but read the license before building a product on it.
Meta released Muse Spark Open, an open-weight model with 816 billion parameters in total and 42 billion active at a time. It handles images natively and has a 1-million-token context window, which is how much text it can hold in mind at once. Meta calls it the most capable US open-weight model so far; that is Meta's own claim, with no benchmark figures listed.
The license is the Muse Model License, which has limited commercial terms, so it is not Apache-style free-for-anything. Meta says the model runs on two Apple M5 Ultra Mac Studios (512 GB of shared memory each) using MLX, Apple's machine-learning toolkit. Hosted versions are on Baseten, Fireworks, Groq, Modal, Nebius, OpenRouter and Together AI; pricing is not stated.
On your desk: Patrick, a 1M-token window on a pair of Macs makes it conceivable to process your whole Obsidian vault locally, though you would want to check the license terms first.
Read the original at Meta Research: Muse Spark Open →