Inception launches Mercury 3, its flagship diffusion LLM, at 1,068 tokens per second

Inception launched Mercury 3 on October 19, 2026, a diffusion language model that generates text in parallel rather than token by token. The company lists a speed of 1,068 tokens per second on NVIDIA GPUs, a 1 million token context window (up from 260,000 in Mercury 2.5), and a model more than 40 times larger than Mercury 1. Launch pricing is $0.10 per million input tokens and $0.35 per million output tokens, half the list price. It is available through Inception, OpenRouter, Baseten and Vercel, and is OpenAI API compatible. Mercury 2 and 2.5 are closed to new users, with existing non-enterprise access ending November 15.

Inception says Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks while running four to eight times faster. Those comparisons are the company's own; independent workloads have not yet settled them. The substantive claim is architectural: that diffusion scaling laws hold like autoregressive ones, so speed no longer has to be traded for capability at the fast model tier.

Sources: Unite.AI; Interesting Engineering; Business Wire