Mercury 3 now available on AI Gateway
Vercel AI Gateway now exposes Mercury 3 to coding agents, pairing diffusion-based generation above 1,000 tokens per second with a 1M-token context window.
Inception’s **Mercury 3** is available through Vercel AI Gateway as `inception/mercury-3`. It uses diffusion-based parallel refinement, with reported generation above **1,000 tokens per second** and a **1M-token context window**.
Builders can test it in Claude Code, Codex, or Cursor after installing the latest Vercel CLI and running setup. It is worth evaluating for agent loops where response latency and long working context matter.
Inception’s **Mercury 3** is available through Vercel AI Gateway as `inception/mercury-3`. It uses diffusion-based parallel refinement, with reported generation above **1,000 tokens per second** and a **1M-token context window**. Builders can test it in Claude Code, Codex, or Cursor after installing the latest Vercel CLI and running setup. It is worth evaluating for agent loops where response latency and long working context matter. The speed and benchmark claims come from Inception, and the material provides no independent coding-agent comparison. The **50% launch discount** is temporary, with no end date stated.
Mercury 3 adds an unusually high claimed-throughput, million-token option for latency-sensitive agent loops, but it does not settle gateway routing. The launch discount and vendor-reported specifications make it a timely evaluation candidate, while coding quality, first-token delay, tool-loop reliability, and sustained performance over long contexts still need matched workload tests.