A 320B sparse-MoE vision-language model with a hybrid linear and sparse-attention backbone and a 1M-token context, running on your Mac out of the box.
Oct 6, 2026→
Download for macOS
