Inception Launches Mercury 2, a Diffusion-Based Reasoning LLM Claiming 5x Speed Over Leading Models

Summary

Inception, the company behind the first commercial diffusion large language models (dLLMs), announced Mercury 2, which it claims is the fastest reasoning LLM available. The company states Mercury 2 is 5x faster than leading speed-optimized LLMs while offering dramatically lower inference costs. The model represents continued development in the diffusion-based approach to language model architecture, which differs fundamentally from the autoregressive token generation used by most mainstream LLMs