Mercury 3 brings diffusion language models to a 1M-token context

What is changing?

Inception released Mercury 3, its flagship diffusion-based language model, on October 19. The company says the model generates 1,068 tokens per second on NVIDIA GPUs, supports a 1-million-token context window and is more than 40 times larger than Mercury 1 by total parameter count. These figures come from the provider’s own tests and announcement.

Mercury 3 is available through the Inception API Platform, OpenRouter, Baseten and Vercel. It uses an OpenAI-compatible API format. Standard pricing is $0.20 per million input tokens and $0.70 per million output tokens; launch pricing lowers those rates to $0.10 and $0.35. Cached input tokens are advertised at a 90% discount.

Rather than producing tokens only from left to right, the diffusion approach refines spans of text in parallel. The architecture targets latency-sensitive workflows including browser agents, real-time search, education and customer support. Teams should validate the provider’s benchmark claims against their own tasks, quality thresholds and total latency costs.