Meta has released Muse Spark Open, an 816B-parameter mixture-of-experts model with 42B active parameters and a 1M-token context window. Built for coding and agentic work, it is natively multimodal. According to Nathan Lambert of Interconnects, the model is now the strongest American-made open-weight model by a considerable margin, and together with Reflection's Beam, it prevents a scenario where more than 20 Chinese open models would outperform the best Western ones.
On performance, Muse Spark Open sits above GLM 5.2 and near DeepSeek V4.1 Flash, while trailing GLM 5.3 and Kimi K3. Lambert estimates the gap between the best American and Chinese open models has been cut roughly in half. He also notes the model appears to be a safety-tuned version of Muse Spark 1.2 rather than a degraded release, as some observers initially suspected.
The main obstacle is the license. It requires products to display "Built with Muse," gates commercial use for companies earning over $20 million per month, and imposes an acceptable use policy that extends to outputs and derivatives, creating legal ambiguity for synthetic data. Internal research is permitted, but Lambert argues the complex, non-standard terms are an unnecessary strategic risk, even if many developers will likely ignore them in practice.