Meta released Muse Spark Open, an 816-billion-parameter open-weight mixture-of-experts model with 42 billion active parameters and 1 million token context window (MarkTechPost). The model processes text, images, video, and audio, and is optimized to run locally on Apple silicon through collaboration with Apple (The Decoder). Performance on open-model benchmarks places it roughly on par with leading open-weight Chinese models (The Decoder), and the model runs on two M5 Ultra Mac Studios (MarkTechPost).
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open multimodal mixture-of-experts agent model with 12 billion active parameters, 1 million token context, and NVFP4 quantized weights that run on a single DGX Spark, plus two compact Bolt models for edge deployment (MarkTechPost). The model targets enterprise agent deployment (The Decoder), with dozens of organizations already fine-tuning it (The Decoder). Separately, Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights image generation model supporting 4K native output, pixel-precise editing, and multi-reference image generation that runs on consumer GPUs with 8 GB of VRAM (The Decoder).