Open weights, very large footprint
Meta’s Muse Spark Open release pairs an open-weight model with a 1-million-token context window and multimodal input. Meta describes the model as 816 billion total parameters, with 42 billion active per token. Those figures explain why “open” does not mean “easy to run on a personal laptop.”
The announcement’s own high-end local demonstration used two Mac Studio systems with M5 Ultra chips. Meta’s license also sets terms for use, so teams should read the actual license rather than treating “open weights” as a synonym for unrestricted use.
A practical way to evaluate it
The most useful question is not only whether the model can accept text and images. It is whether its context length, quality, latency and memory requirements fit the task you want to run. Try a representative multimodal workflow, then measure end-to-end time and hardware cost against a hosted option.
Meta’s specifications and examples are manufacturer-reported. The 1-million-token window is a maximum context capability, not a guarantee that every application can use that much context efficiently.
Source published 2026-10-15. Coverage is based on the maker’s announcement and demonstration.