Inception released Mercury 3, claiming speeds above 1,000 tokens per second, a one-million-token context window and benchmark results ahead of two competing models.
Inception announced Mercury 3 on October 19, introducing a new version of its diffusion-based language model. The company says the model is available through its API and through OpenRouter, Baseten and Vercel.
Inception reports a generation speed of 1,068 tokens per second on NVIDIA GPUs and a context window of one million tokens, up from 260,000 for Mercury 2.5. The company says Mercury 3 performed better on a range of benchmarks than GPT-6 Luna and Claude Haiku 5.5 while running four to eight times faster. Those performance comparisons are company-reported; the announcement does not establish that the results generalize to every task.
The listed price is $0.20 per million input tokens and $0.70 per million output tokens, with a launch discount reducing those rates by half. Inception also says early-access partners are using the model for voice and chat support, tutoring, data engineering and web navigation. Its API is described as OpenAI-compatible, and the company offers enterprise deployments with dedicated capacity and configurable data-retention controls.
Mercury 3 is notable as another commercial effort to scale diffusion beyond image generation into language-model use. Unlike standard autoregressive models, which produce text one token at a time, diffusion models generate and refine text through a different process. Inception’s speed figures and benchmark comparisons will be of interest to developers considering latency-sensitive applications, but independent evaluations will help establish how the model performs across workloads and deployment conditions.
The model is available now through the listed API and distribution platforms. Inception says it plans to continue expanding its real-time AI product portfolio, which also includes Mercury Voice and Mercury Decide.