Meta launches 816B Muse Spark Open and closed Wonder models

Meta released Muse Spark Open on October 15: an 816-billion-parameter mixture-of-experts with 42 billion active per token, a one-million-token context window, natively multimodal, and "built for coding and agentic work." Alongside it came two closed-weight frontier models, Muse Wonder and Muse Wonder Full. On Interconnects, researcher Nathan Lambert called Spark Open "the strongest American-made open model by quite some margin."

Lambert places Spark Open "somewhere above GLM 5.2, near DeepSeek V4.1 Flash, and behind GLM 5.3 or Kimi K3." The closed Wonder tier he ranks "only behind Anthropic's Opus 5.5 and similar to OpenAI's GPT-6-Astra or Google's Gemini-4-Argon." His framing of the week: "Meta is the only American company with frontier open and closed models."

The license is the catch. Users must display "Built with Muse" on their products. "Companies that make more than $20M per month across their entire business deploying the model need a separate license," Lambert writes, with an acceptable use policy and restrictions on synthetic data use rounding it out. He calls the custom license "an unnecessary strategic risk," while noting it does permit internal research and applications.

Lambert's read on the shape of the field: "The costs of building true frontier models are now measured in billions per major model." Without this week's drop, by his count, "20+ open-weight models built by Chinese companies" would sit above anything American. On X he put it more plainly: "I waited so long for US open weight models to be relevant again."