Mercury 2 For Search

Mercury 2 targets the waiting time that accumulates inside a search assistant. Rewriting queries, ranking retrieved passages and composing an answer can each require another model call. Inception says its diffusion model refines a span of text in parallel over successive passes, reducing delay at each step. In its 100-question FRAMES test, the company reports 10.8 seconds per query. Faster intermediate calls could leave room for more search work within a response-time budget; the company's tests also show that speed and answer completeness remain separate considerations.