Meta opens weights on 816B-parameter Muse Spark model

Meta released open weights for Muse Spark Open, an 816-billion-parameter model built for coding and long-running agent work. Only 42 billion parameters activate per token, and the context window reaches 1 million tokens with support for text, images, audio, and video inputs. Meta calls it the most capable open-weight model from a U.S. lab to date. On the JobBench agent test it scored 61.8, more than double rival Inkling's 31.1, though Chinese models led on coding benchmarks. A new license allows free commercial use but requires a separate agreement from hosted-service providers exceeding $20 million in annual revenue.