TL;DR: Meta's Muse Spark Open ships as the largest US open-weight model to date, pairing a 1M-token context with native multimodal input and a sparse 816B/42B parameter design.
Summary: Meta AI released Muse Spark Open, which it describes as the most capable US open-weight model to date. The model uses 816B total parameters with 42B active per token — a sparse mixture-of-experts-style layout — and supports a 1M-token context window, with native multimodality across text, image, video, and audio. Meta positions it specifically for coding and agentic workloads, claiming strength on long-horizon tasks, complex software engineering, and multi-step reasoning.
Why it matters: A frontier-scale open-weight model at 1M context with native multimodal input gives builders a self-hostable or fine-tunable alternative for long-horizon agent and coding pipelines. Worth benchmarking against your current agent stack on multi-step repo-level tasks, and checking the license and serving cost implied by 816B total parameters.
Source: x_com