Inception's Mercury 3 proves diffusion LLMs can scale, hitting 1,000 tokens per second
Inception "Mercury 3", 1000 tok/s diffusion LLM
Inception's Mercury 3 is a diffusion-based LLM that outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4–8x faster at over 1,000 tokens per second on NVIDIA GPUs. It's more than 40x larger than Mercury 1 yet just as fast, with a 1M-token context and pricing at $0.20 per million input tokens. Early partners use it for voice support, tutoring, search agents, and web navigation.
The scaling laws that drove progress in autoregressive models hold for diffusion too.