Mercury 3 brings a 1M context diffusion LLM to hosted APIs
Inception released Mercury 3, a diffusion language model with a one-million-token context window, OpenAI-compatible API, list pricing of $0.20 per million input and $0.70 per million output tokens, and a launch discount at half those rates. Inception reports 1,068 tokens per second on NVIDIA GPUs and distribution through its API, OpenRouter, Baseten and Vercel.
Why it matters: A fast diffusion model could shorten multi-step agent loops and voice or search latency, but the practical decision is whether speed survives the exact provider, prompt length and accepted-quality threshold.