DiffuSpace Raises Hundreds of Millions of RMB, Largest-Ever Funding for Diffusion Language Models

DiffuSpace, a startup focused on diffusion language models, announced on October 14 that it has completed a funding round of several hundred million RMB, described as the largest ever for the diffusion language model direction. The round was co-led by Matrix Partners China, Shunwei Capital, and Legend Capital, with participation from Zhongke Chuangxing, Huawei Hubble, and Horizon Robotics.

Founded by Kong Lingpeng, a professor at the University of Hong Kong, DiffuSpace is pursuing a diffusion-based approach to language modeling, which differs from the autoregressive architecture used by mainstream large models. Diffusion models generate text in parallel and allow edits across any position rather than generating token by token. The company also said it plans to release a larger model in October 2026.

Shunwei Capital was founded by Lei Jun, and Huawei Hubble is the investment platform under Huawei. While mainstream large models predominantly use autoregressive architectures, diffusion language models are seen as one potential direction of technological change, with companies such as Google also exploring the space.