Mercury 3
Inception Labs · 2026-10-19
Mercury 3 is Inception Labs' October 19, 2026 diffusion language model for agentic and latency-sensitive text workloads, served as mercury-3 on the OpenAI-compatible Inception API and through OpenRouter, Baseten and Vercel. Inception's model documentation lists tool calling, up to 131,072 output tokens and instant, low, medium and high reasoning effort, and advertises a 1M-token context, up from 260K for Mercury 2.5. Inception says Mercury 3 has more than 40 times the total parameters of Mercury 1, but publishes no exact parameter count or unrounded context limit.