(Image source: unsplash)
TMTPOST -- A groundbreaking collaboration between Associate Professor Dai Jifeng, a leading AI scientist at Tsinghua University, and Chen Tianqiao, the founder of Tianqiao and Chrissy Chen Institute and an innovative entrepreneur and philanthropist, has put a new spotlight on China’s ambitions in artificial intelligence.
Their new venture, aimed at advancing Artificial General Intelligence (AGI), officially debuted on Sunday with the launch of the MiroMind Open Deep Research (Miro ODR) project, a high-performance, fully open-source AI research platform that aims to challenge global leaders in the field.
The MiroMind AI team, under Dai’s direction, revealed the initial version of Miro ODR, which achieved an impressive 82.4 score on the GAIA benchmark test. This performance surpassed not only OpenAI’s DeepResearch and Manus but also other prominent open-source and proprietary AI deep research models, positioning it as the strongest open-source model available from Monday.
The project, which includes four key subcomponents, is now accessible on popular open platforms such as GitHub and Hugging Face, signaling a commitment to transparency and collaborative progress in AI development.
Dai emphasized that the project is fully reproducible, with open access to core models, datasets, training methodologies, AI infrastructure, and the deep research agent framework. The team plans monthly updates and encourages the AI community to contribute toward building the most powerful deep research model possible.
Chen, whose backing provides critical support, has reportedly pledged that half of the profits generated by AI companies incubated within his enterprise Shanda will be shared with the MiroMind team. “At MiroMind, we don’t provide AI—we build AI together with you,” Dai stated, underscoring the collaborative ethos of the initiative.
Dai’s leadership brings significant experience and expertise to this endeavor. He completed his bachelor’s and doctoral studies in engineering at Tsinghua University and held influential research positions at Microsoft Research Asia and SenseTime Research Institute before returning to Tsinghua as an associate professor.
His research focuses on foundational models and algorithms for visual information processing, an area in which he has published extensively. With more than 80 papers and over 60,000 citations, Dai’s work has influenced both academia and industry worldwide.
His contributions include the development of the Deformable ConvNets series, which has been incorporated into computer vision courses at top universities and used in key deep learning frameworks, often outperforming offerings from global tech giants like Microsoft, Meta, and Google.
Dai’s R-FCN object detection algorithm is recognized as a foundational model in the field, and his BEVFormer framework has attracted attention from leaders such as NVIDIA CEO Jensen Huang. Additionally, Dai has contributed to multimodal foundational models, including InternVL, an open-source model that rivals commercial closed-source systems like OpenAI’s GPT and Google’s Gemini, having been downloaded over 5 million times.
MiroMind’s technical report highlights how the project differs from existing deep research approaches by opening every stage of the research process. The four subprojects—MiroFlow, MiroThinker, MiroVerse, and MiroTrain—cover the AI research lifecycle and are even optimized for mobile devices. MiroFlow acts as an agent framework that integrates multiple tools to extend large language models and enables tool-assisted deep reasoning, achieving top-tier GAIA scores.
MiroThinker is a deep reasoning model designed for native tool-assisted logic, achieving state-of-the-art results on benchmark datasets. MiroVerse provides an open data repository with 147,000 training points, continuously updated to support research. MiroTrain offers the infrastructure for stable and efficient training, covering workflows including long-text and reinforcement learning training.
Central to these efforts is MiroMind-M1, an open-source reasoning language model based on Alibaba’s Qwen-2.5 architecture with seven billion parameters. This model specializes in mathematical reasoning and has been trained using a combination of supervised fine-tuning on hundreds of thousands of questions and reinforcement learning with verifiable rewards. Its performance on leading mathematical benchmarks has surpassed competitors like DeepSeek-R1 and Xiaomi’s MiMo-7B-Base, with improved efficiency demonstrated through fewer tokens required for comparable results.
Dai has shared that MiroMind’s inference framework outperforms many closed-source alternatives and that the core deep research model is entirely self-developed and open source. The team recently opened the MiroMind ODR demo for public trial, inviting broader participation and feedback.
In addition to foundational AI research, the startup is focusing on practical applications in AI-driven business decision-making, breaking content distribution algorithmic echo chambers, and developing AI services tailored for aging populations and youth. The company’s mission statement emphasizes building digital lifeforms with self-awareness, evolving alongside the community to ensure the development of safe and beneficial AGI.
Parallel to the AI advances spearheaded by Dai, Chen is accelerating investments in neuroscience and brain-computer interface (BCI) technologies. His backing of Synchron, a U.S.-based vascular interventional BCI company, led to a recent milestone demonstration where an ALS patient controlled an Apple iPad entirely through thought. This represents the first-ever native mind control on an Apple device and is considered a breakthrough in non-invasive brain-machine interaction technology.
Chen’s investment in Synchron predates those of other notable tech figures like Bill Gates and Jeff Bezos. Unlike invasive BCI methods, which involve implanting chips directly into the brain, Synchron’s technology leverages EEG and MRI to non-invasively capture neural signals, offering a safer and more advanced alternative. Chen believes that AI, combined with deep data analytics and machine learning, can achieve or even surpass the capabilities of invasive methods without risking brain damage.
Since 2016, Chen and his wife Chrissy Luo have committed over $1 billion to brain science research globally. Their support has enabled initiatives ranging from the Tianqiao and Chrissy Chen Institute for Neuroscience at Caltech to partnerships with Chinese medical institutions, fostering research and building extensive neural data resources. Chen has also been a vocal advocate for long-term, patient capital investment in hard technology sectors like AI and neuroscience, emphasizing the need for sustained funding and deep industry understanding over the quick-return mentality common in internet startups.
The Tianqiao and Chrissy Chen Institute recently launched the AI-Driven Science Award in collaboration with the American Association for the Advancement of Science, aimed at recognizing young scientists who use AI to advance scientific discovery. The award provides an international platform for emerging researchers to gain recognition and promote innovation in artificial intelligence.
