Inception says its Mercury 3 model beats GPT-6 Luna and runs far faster

Inception launched Mercury 3, a diffusion-based large language model that reportedly outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4-8 times faster and handling up to one million tokens at once. The model's pricing is competitive, emphasizing speed for scalable applications like agents and coding tools. Inception also used Mercury 3 agents in a recursive process to enhance the model's speed by 16%.