Diffusion at Scale: Introducing Mercury 3
Summary
Mercury 3 is a diffusion-based LLM released by Inception, boasting 1,068 tokens per second on NVIDIA GPUs and a 1M token context. The article presents Mercury 3 as performing on par with or better than frontier models like GPT-6 Luna and Claude Haiku 5.5, with real-world production use cases in browser agents, interactive education, real-time search, and customer support, plus availability via multiple API channels and OpenAI API compatibility.