Soumye Singhal

Multimodal models can now handle audio natively with improved efficiency, achieving state-of-the-art results in complex tasks like document understanding and agentic computer use.