Mercury 2: Fast reasoning LLM powered by diffusion | MakersHub News