Inception Labs builds Mercury, the first family of commercial-scale diffusion large language models (dLLMs) that generate tokens in parallel for 5-10x faster inference than comparable speed-optimized models. - View it on GitHub