DiffuSpace reportedly completes two financing rounds totaling nearly 500000000 yuan, plans to train and open-source a 30000000000-parameter model
BIBIBI
AT A GLANCE
DiffuSpace reportedly completes financing of nearly 500000000 yuan and plans to release an open-source 30000000000-parameter model.
Article
HKU diffusion model team raises nearly 500000000 yuan, with Huawei and Shunwei betting on it
Beating AI flash: Shenzhen-based diffusion language model company DiffuSpace has consecutively completed two financing rounds, with total funding approaching 500000000 yuan. Matrix Partners China, Shunwei Capital, and Legend Capital jointly led the rounds, with Huawei Hubble and Horizon Robotics among the participating investors. Multiple media outlets described this as the largest financing deal globally in the diffusion language model field.
DiffuSpace was founded this year by May Lingpeng Kong, a professor at the University of Hong Kong, and his doctoral students Shansan Gong and Jiacheng Ye. The team previously launched Dream 7B, and has open-sourced the model weights. It uses diffusion language model (dLLM) technology.GPT, Claude, and other mainstream models usually generate text sequentially, whereas diffusion models can process content at multiple positions simultaneously and then progressively complete and modify it.
DiffuSpace is training a new model with 30000000000 parameters, which it plans to release and open-source in the near term. The company will focus on advancing code Agents and local deployment on devices such as phones, cars, and robots.
Original link https://m.theblockbeats.info/flash/370963
Key points
01
Beating AI says that DiffuSpace has consecutively completed two financing rounds totaling nearly 500000000 yuan.
02
The report names Matrix Partners China, Shunwei Capital, and Legend Capital as the lead investors, and Huawei Hubble and Horizon Robotics as participating investors.
03
DiffuSpace was founded this year by May Lingpeng Kong, a professor at the University of Hong Kong, and his doctoral students Shansan Gong and Jiacheng Ye.
04
The team previously launched Dream 7B jointly with Huawei Noah's Ark Lab and has open-sourced the model weights.
05
DiffuSpace is training a new model with 30000000000 parameters, which it plans to release and open-source in the near term.
06
The company will focus on advancing code Agents and local deployment on devices such as phones, cars, and robots.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
DiffuSpace is a company that uses diffusion language model technology. The report says it has completed two financing rounds and is preparing to train and open-source a larger model while exploring code tools and on-device operation.
Why it matters to readers
If the reported content is confirmed, this would indicate that diffusion language models have received financial support from multiple institutions and companies and may continue expanding into code and device-side applications.
You can understand it this way: the team has obtained funding to build a larger model and run it on devices such as phones, cars, and robots.
Risks and unknowns
The current information comes from a single reposted flash and has not yet been accompanied by a company announcement, investor confirmation, or financing documents.
The financing amount, participating investors, and the claim that this is “the largest financing deal globally in the diffusion language model field” have not been independently verified.
30000000000 The release date, open-source address, performance evaluations, and actual deployment progress of the parameter model remain unclear.
Related Developments
Loading event timeline…
Related concepts
Diffusion language model (dLLM)
This term is not in the glossary yet. Browse related concepts in the glossary.
