Multilingual, multimodal LLMs (30B and 235B) with advanced reasoning, agent capabilities, and seamless mode switching.