Meta发布开源Muse Spark Open模型，原生支持Mac运行

Meta launches Muse Spark Open: open-weight AI model optimized for Macs

Muse Spark Open is Meta's first open-weight model explicitly engineered for on-device execution on Apple Silicon—running efficiently on M-series Macs without cloud dependency. Unlike previous open models requiring heavy quantization or custom runtimes, it ships with native Metal acceleration and memory-aware inference, enabling real-world app integration out of the box. While parameter count remains unspecified, its 'US-based' labeling and Apple co-optimization signal a strategic shift toward sovereign, deployable AI infrastructure. For early AI teams: this lowers the barrier to shipping performant, privacy-preserving applications on consumer hardware. Founders should prioritize evaluating Muse Spark Open for client-side logic—replacing API calls with local inference where latency, cost, or compliance matter. It signals that 'device-native AI' is no longer aspirational but immediately actionable—and that choosing the right open model now means checking not just capability, but actual runtime compatibility.