AitopkitInception, the company behind the first commercial diffusion large language models (dLLMs), announced Mercury 3 on October 19, 2026 via Business Wire, calling it the most capable diffusion LLM ever released. The headline numbers are striking: the model is more than 40 times larger than Mercury 1, yet still runs at over 1,000 tokens per second — Inception reports 1,068 tokens per second on NVIDIA GPUs.
A diffusion LLM generates text by gradual refinement, the same technique image diffusion models use, instead of the token-by-token prediction of autoregressive models. Inception frames Mercury 3 as proof that this approach scales: intelligence tripled across standard benchmarks from Mercury 1 (mini) to Mercury 3, and the company says the same scaling laws that drove autoregressive progress hold for diffusion too. The context window also jumps from 260,000 tokens in Mercury 2.5 to 1 million tokens.
Inception claims Mercury 3 outperforms GPT-6 Luna and Claude Haiku 5.5 on a broad range of standard benchmarks while running 4–8x faster, and topped those models on 8 of 12 benchmark suites in its own comparison. Launch pricing is $0.10 per million input tokens and $0.35 per million output tokens (list price $0.20/$0.70, with cached input priced 90% below input), available through the Inception API Platform, OpenRouter, Baseten and Vercel. Early customers are already running it in browser agents, interactive education, real-time search agents and customer support.
As with every vendor benchmark claim, independent verification is still pending — but if Mercury 3 holds up, it would break the long-standing trade-off between speed and intelligence, and give developers a compelling option in the ongoing AI price war. For a broader look at how fast, cheap models compare to the big names, see our ChatGPT vs Claude comparison.