NVIDIA researchers published an arXiv preprint in July 2026 introducing Nemotron-Labs-Diffusion, a language model family available in 3B, 8B, and 14B parameter sizes, downloadable from Hugging Face under a license that permits commercial use. The models operate in three distinct generation modes from a single set of weights, switching by changing only the attention pattern. The core result, according to the NVIDIA-authored preprint, is that the 8B instruct variant decodes six times more tokens per forward pass than Qwen3-8B in self-speculation mode and translates that to four times higher throughput on SPEED-Bench when served with SGLang on a GB200 GPU. The preprint has not yet undergone peer review or independent replication. The approach replaces one-token-at-a-time autoregressive generation with block-level parallel denoising, addressing the memory-bandwidth-bound regime of single-token LLM inference.