AitopkitMeta has released Muse Spark Open on October 15, 2026 — an 816-billion-parameter open-weight mixture-of-experts model from Meta Superintelligence Labs, with 42 billion parameters active per token and a 1-million-token context window.
Meta calls it "the most capable US open-weight model to date" — a direct shot at the Chinese labs that have dominated open-weight leaderboards this year. Day-one distribution is unusually wide: seven hosted-inference providers and four weight-distribution channels, including downloadable weights on Hugging Face. Serverless API pricing starts at $1.25 per million input tokens and $4.25 per million output tokens on providers like Together AI.
The model accepts text, images, video, and audio as input and generates text output — a true multimodal frontier release. It ships under Meta's custom Muse Model License, which sets the terms for commercial use of the weights.
For developers, the pitch is frontier-class capability without API lock-in: a model you can self-host or run through your preferred cloud, with the open-weight ecosystem (fine-tuning, distillation, on-prem deployment) that closed APIs don't offer. The caveat is the usual one: "open-weight" is not open-source — read the license before building on it.
Muse Spark Open lands in the most crowded open-weight month of 2026. If your team was waiting for a US answer to DeepSeek and Qwen, this week finally delivered one.