Meta releases MuseSpark Open with first-party Apple hardware support
Meta launched MuseSpark Open, an 816 billion parameter model optimized for Apple M5 Ultra chips. Nvidia and Black Forest Labs also released new open-weight models for enterprise agents and image generation.
Meta released MuseSpark Open, an 816 billion parameter open-weight model with 42 billion active parameters and a one-million token context window. AI CEO Alexander Wang stated the model is built for coding and agentic work. Artificial Analysis scored the model at 39, placing it 17th overall and 7th among open models. It trails Kimi K3 slightly on coding benchmarks but claims state-of-the-art results on agentic benchmarks such as Jobbench and MCP Atlas. The release marks Meta's return to top-tier open-weight models following the underperformance of Llama 4.
Apple provided first-party support by updating its MLX software toolkit to ensure the model runs on Apple hardware. The full model requires a terabyte of memory, necessitating a pair of M5 Ultra Mac Studios costing approximately $40,000. Apple CEO John Ternus and Meta CEO Mark Zuckerberg publicly praised the collaboration, with Wang crediting former Apple executive Daniel Gross for smoothing past grievances. Meta and Apple are positioning this partnership as a significant step for on-device AI, with Wang noting the model runs efficiently on Mac Studios.
In related releases, Nvidia launched Nemotron 3.5 Super VL, a 120-billion parameter open model distilled from Nemotron 3 Ultra. It is designed for enterprise agents to run on single GPUs with five times higher token throughput than similar open models. Black Forest Labs released Flux 3 Image Klein, a 5.6-billion parameter image generation model that runs on 8GB GPUs and offers free commercial use for companies with under $5 million in annual revenue.