Former Shanghai AI Lab large-model head founds SuperEvolution Intelligence, completes several hundred million yuan in angel-round financing
BIBIBI
AT A GLANCE
SuperEvolution Intelligence reportedly completed several hundred million yuan in angel-round financing and plans to develop foundation models capable of conducting research autonomously and continuously improving.
Article
Former Shanghai AI Lab Large-model lead raises hundreds of millions of yuan to enable AI itself to conduct research and improve its models
Dongcha Beating AI News Flash, former Shanghai AI Lab large-model center head Chen Kai's founded company, Super Evolution Intelligence, recently completed a several-hundred-million-yuan angel round led by BlueRun Ventures, with participation from SenseTime Guoxiang Capital, Lenovo Star and other institutions. The company only began formal operations at the end of this year August and aims to build foundation models capable of conducting research autonomously and continuously improving their own capabilities.
Chen Kai previously led the research and development of the InternLM series of large models and has experience training models with trillions of parameters. In this venture, however, he does not plan to begin by training a trillion-parameter model from scratch. Instead, he will first perform post-training based on open-source models, starting at the scale of hundreds of billions of parameters, and then gradually explore mid-training and pre-training.
The company is currently building a self-evolving training loop, preparing research data and validation environments, and establishing a corresponding evaluation system. Chen Kai plans to launch the first model using the new paradigm before mid-2027 year.
Original link https://m.theblockbeats.info/flash/374080
Key points
01
SuperEvolution Intelligence was founded by Chen Kai, former head of the Shanghai AI Lab large-model center.
02
The report states that SuperEvolution Intelligence recently completed several hundred million yuan in angel-round financing.
03
The funding round was led by BlueRun Ventures, with participation from SenseTime Guoxiang Capital, Lenovo Star and other institutions.
04
The company aims to build foundation models capable of conducting research autonomously and continuously improving their own capabilities.
05
Chen Kai said the company will first perform post-training based on open-source models, starting at the scale of several hundred billion parameters, and then gradually explore mid-training and pre-training.
06
The company is building a self-evolving training loop and preparing research data, a validation environment, and an evaluation system.
07
Chen Kai plans to launch 2027 the first model adopting the new paradigm before mid-year.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
The company hopes to first continue training open-source models rather than immediately training a trillion-parameter model from scratch; at the same time, it is building data, validation, and evaluation processes to explore enabling models to participate in research and continuously improve their own capabilities.
Why it matters to readers
The report also disclosed the early-stage financing, technical roadmap, and model launch plan, providing clear milestones for observing whether the project can translate its goal of “autonomous research and self-improvement” into a model.
Beginners should distinguish between reported financing that has already occurred and the R&D goals and 2027 product plan that have not yet been realized.
Risks and unknowns
The original article did not provide a formal financing announcement issued by the company or participating investment institutions, and the financing amount and participation details lack independent confirmation.
“Conducting research independently” and “continuously improving its own capabilities” are currently company objectives; their actual effectiveness has not yet been validated.
It remains uncertain whether the first model can be launched before mid-year 2027.
The original text does not specify which year “this year August end” refers to.
The source text does not disclose the specific financing amount, valuation, or investment proportion.
Related Developments
Loading event timeline…
Related concepts
Post-training
This term is not in the glossary yet. Browse related concepts in the glossary.
