Inception introduced Mercury 3, a diffusion-based language model featuring a 1-million-token context window.
Inception introduced Mercury 3, a diffusion-based language model featuring a 1-million-token context window.
Unlike traditional language models, Mercury 3 generates text in parallel, achieving 1,068 tokens per second and running 4–8 times faster than models such as GPT-6 Luna and Claude Haiku 5.5, according to the company's benchmarks. The model is available through Inception API, OpenRouter, Baseten, and Vercel.