Multilingual, multimodal LLMs (0.6B–32B) with advanced reasoning, agent capabilities, and seamless mode switching.