Inception "Mercury 3", 1000 tok/s diffusion LLM
Developers on Hacker News discussed a new language model called Mercury 3, which employs diffusion-based generation. The model reportedly processes roughly 1,000 tokens per second, offering faster output than many existing LLMs. Mercury 3’s architecture aims to improve efficiency while maintaining text quality, and its release has sparked interest across the AI research community for potential applications.