Applications
Mercury 3: Inception scales up its language model to over 1,000 tokens per second
Forty times larger than the first Mercury but just as fast, Inception’s new diffusion model claims to outperform GPT-6 Luna and Claude Haiku 5.5.