Inception Mercury 3 claims up to 8x speed over traditional AI
Inception Mercury 3, announced October 19, uses a diffusion approach to produce over 1,000 tokens per second, though its benchmark claims are not yet independently verified.
Inception released Mercury 3 on October 19, 2026, announced by co-founder Stefano Ermon.
The company says the diffusion-based model generates text up to eight times faster than traditional AI models.
Inception says Mercury 3 produces over 1,000 tokens per second on widely available NVIDIA GPUs.
The context window rises to 1 million tokens from 260,000 in Mercury 2.5.
Inception claims it beats GPT-6 Luna and Claude Haiku 5.5, but has not published benchmark names or scores.