AI lab pioneering diffusion large language models (dLLMs) with the Mercury family for fast parallel-token generation.