tl;dr: Inception's Mercury 3 is a diffusion-based LLM that scales 40x beyond Mercury 1, claiming parity with GPT-6 Luna and Claude Haiku 5.5 on benchmarks while running 4–8x faster at 1,000+ tokens/sec. It supports a 1M token context, costs $0.20/$0.70 per million input/output tokens (50% off at launch), and is being used by early partners for browser agents, tutoring, search, and customer support. The release argues that autoregressive scaling laws also hold for diffusion models.