Article 282: Meta's Muse Spark Open Puts an 816-Billion-Parameter Model on Two Macs

Meta has released Muse Spark Open, an 816-billion-parameter open-weight mixture-of-experts model with a 1M-token context window, pitched at coding, long-horizon agentic tasks and multi-step reasoning. Meta's AI chief Alexandr Wang called it "the most capable US open-weight model to date," positioning it directly against Alibaba's Qwen, Moonshot's Kimi K3 and Z.ai's GLM, which have dominated the open-weight field. Unusually, Meta says the model is optimised for Apple's M5 Ultra chips: two Mac Studios can run it as a desktop "supercomputer."

Key Takeaway: Read the distribution choice, not the benchmark. Meta spent two years as the open-weight leader, then pivoted to closed models to fight OpenAI and Anthropic - and is now returning to open weights with the flagship of its lineup. Doing that on Apple silicon rather than through a hyperscaler is a deliberate bet that the constraint has moved. If the strongest US open-weight model ships pre-tuned for hardware already sitting on desks, the scarce commodity is no longer access to a model but the physical footprint, power and cooling of the data centre. That is a very different competitive map from the one everyone has been drawing.

Sources: Meta AI, Reuters, The Guardian, TechCrunch, MacRumors, 15-16 October 2026