Inception has released Mercury 3, its most capable diffusion-based language model to date, designed to compete directly with cost-optimized frontier models like GPT-6 Luna and Claude Haiku 5.5. Unlike conventional autoregressive models that generate text one token at a time in sequence, Mercury 3 utilizes a diffusion process to produce and refine multiple tokens simultaneously in parallel. The model delivers exceptional performance, achieving speeds of over 1,000 tokens per second on standard NVIDIA GPUs while supporting a 1-million-token context window. Priced aggressively at $0.10 per million input tokens and $0.35 per million output tokens, the model significantly slashes infrastructure and inference costs for real-time applications. Industry analysts highlight that such efficiency breakthroughs in diffusion architecture could fundamentally disrupt the economics of high-frequency agentic workflows, voice pipelines, and search tools.

Full report : By generating tokens in parallel rather than sequentially, Mercury 3 achieves over 1,000 tokens per second at a fraction of standard operational costs.