Inception’s Mercury 2 speeds around LLM latency bottlenecks

Inception has introduced Mercury 2, calling it the world’s fastest reasoning LLM. Intended for production AI, the large language model leverages parallel refinement rather than sequential decoding.

Mercury 2 was announced February 24, with access requests available on Inception’s website. Developers can also try Mercury 2 using the Inception chat.

Inception says Mercury 2 is intended to solve a common LLM bottleneck involving autoregressive sequential decoding. The model instead generates responses through parallel refinement, a process that produces multiple tokens simultaneously and converges over a small number of steps, Inception said. Parallel refinement results in much faster generation and also changes the reasoning trade-off, according to the announcement. Higher intelligence typically leads to more computation at test time, meaning longer chains, more samples, and more retries. This all results in higher latency and costs. Mercury 2 uses diffusion-based reasoning to provide reasoning-grade quality inside real-time latency budgets, said the company.

Mercury 2 is OpenAI API-compatible and especially suited to latency-sensitive applications where the user experience is non-negotiable, the company said. Use cases include coding and editing, agentic loops, real-time voice and interaction, and pipelines for search and RAG operations.