Inception released Mercury 3 for teams building voice agents, live tutoring, browser agents, and customer support, where waiting on the model is visible to the user. The model generates text by diffusion, producing tokens in parallel. Inception measures 1,068 tokens per second on NVIDIA GPUs and says Mercury 3 matches or beats GPT-6 Luna and Claude Haiku 5.5 on quality while running 4 to 8 times faster. The release is Inception's case that diffusion language models can grow in size and keep that speed. The company says Mercury 3 is more than 40 times larger than Mercury 1, with scores on a range of standard benchmarks about 3 times higher than Mercury 1 mini, at similar throughput. Context is 1 million tokens, up from 260,000 on Mercury 2.5. List price is $0.20 per million input tokens and $0.70 per million output tokens, 50 percent off at launch, and cached input is 90 percent below the input rate. Early users at TinyFish, RevisionDojo, Octen, and Lorikeet are already running it for web navigation, tutoring, search, and support. Mercury 3 is available on the Inception API and through OpenRouter, Baseten, and Vercel. The API is OpenAI-compatible and supports tool calling, structured outputs, and reasoning settings from instant to high. New API accounts start with 100 million free tokens. Inception was co-founded by Stanford professor Stefano Ermon, who is CEO, with UCLA professor Aditya Grover and Cornell professor Volodymyr Kuleshov, to move diffusion methods from images and video into production language models.