Mianbi Intelligence, a Tsinghua University-affiliated AI company focusing on edge-side large language models, has recently secured a new funding round worth hundreds of millions of yuan.
APPSO learned that this round was jointly led by Loongson Capital, DH Fund, Zhongguancun Science City Fund, and SAIF Partners, with follow-up investments from Beijing AI Industry Investment Fund and ChainValley Capital. Wanjia Capital served as the exclusive financial advisor for this round.
Following this funding round, Mianbi Intelligence stated that the company will further accelerate its commercial deployment of efficient large language models represented by “edge-side AI,” aiming to deeply serve industries with higher performance, lower energy consumption, and faster speed at equivalent parameters, creating “tangibly perceivable” value for users.
Mianbi Intelligence is a large language model company that believes in “efficiency as the first principle,” with efficiency being the team’s inherent DNA. The Mianbi team reportedly began large language model research ahead of all other domestic teams and was among the first to implement trillion-parameter multimodal models, providing globally leading edge-side intelligent solutions.
In terms of business, the Mianbi team has made significant and rapid progress in multiple fields. They have not only integrated mainstream consumer electronics and emerging hardware with edge-side AI but have also made achievements in vertical fields such as law and education, benefiting from their technical background in natural language processing and deep accumulation of quality industry data.
Mianbi Intelligence CEO Li Dahai stated that the large language model competition has entered a “protracted war” phase. Additionally, he believes Mianbi Intelligence will maintain its competitiveness in the edge-side intelligence field and is establishing its own market position.
Founded in August 2022, Mianbi Intelligence’s core research members come from Tsinghua University’s NLP Lab and are among the first teams in China to work on pre-trained large language models. To date, they have launched several AI large language model products:
- November 14, 2023: Mianbi’s multimodal large language model application “Mianbi Luca” officially opened to the public
- November 15, 2023: Launched “Mianbi ChatDev,” an AI-native application development platform based on swarm intelligence, while upgrading their CPM series foundation models to the new trillion-parameter large language model “Mianbi CPM-Cricket”
- February 2024: Released open-source edge-side model “MiniCPM,” a key step in their “large model + intelligent agent” dual-engine strategy
- September 2024: Launched “MiniCPM 3.0,” the latest product in the “Little Canon” series
Notably, in July 2023, Mianbi Intelligence, People’s Court Press, and Shenzhen Dibo jointly supported Shenzhen Intermediate Court in officially launching the nation’s first “judicial trial vertical domain large language model.” In November, the “Faxin Legal Foundation Large Language Model,” which Mianbi Intelligence participated in as a joint development team, was released by the Supreme Court, positioned as a national-level legal AI infrastructure.
Furthermore, Mianbi Intelligence is closely collaborating with industry benchmark enterprises such as Huawei, MediaTek, Lenovo, Intel, Great Wall Motors, and Yilai Intelligence, covering areas including AI Phone, AIPC, smart cockpits, smart homes, and embodied robots.
Mianbi Intelligence expects edge-side models reaching GPT-4 level capabilities to be achieved by the end of 2026.
After completing this funding round, Li Dahai issued a company-wide letter. The full text follows:
Dear Mianbi Intelligence colleagues,
I am delighted to share with you that Mianbi Intelligence has completed a new funding round of hundreds of millions of yuan, jointly led by Loongson Capital, DH Fund, Zhongguancun Science City Fund, and SAIF Partners, with Beijing AI Industry Investment Fund and ChainValley Capital as follow-up investors.
With this funding round secured, we stand on a new level: we will further accelerate our commercial deployment of efficient large language models represented by edge-side AI, deeply serving industries with higher performance, lower energy consumption, and faster speed at equivalent parameters, creating tangibly perceivable value for users.
Over the past year, our achievements speak for themselves. Our Little Canon edge-side model has gained global recognition and established a firm foothold. Our commercial deployment and business development based on efficient large language models and edge-side AI are flourishing and progressing rapidly.
With a small, lean startup team, we’ve pushed MiniCPM to edge-side ChatGPT and GPT-4V moments; partnered with industry leaders like Huawei, MediaTek, Lenovo, Intel, Great Wall Motors, Wutong Technology, and Yilai Intelligence to become an indispensable player in industry innovations including AIPC, AIPhone, smart cockpits, and embodied robots; and deeply participated in building the national-level AI infrastructure “Faxin Legal Foundation Large Language Model.”
Today, within China’s “6+2” large model landscape, we’ve become a highly distinctive and competitive representative. Globally, Mianbi is also a benchmark in edge-side intelligence, making it onto Fortune’s Global AI Innovation Top 50!
This is a victory for the Little Canon spirit! We deserve to be proud!
Same time frame, more forward-looking judgments; same resources, more explosive creativity; same opportunities, determination to solve industry and customer pain points thoroughly.
Pushing efficiency to the extreme and creating results beyond expectations, our confidence comes from our deep understanding of large model fundamental principles and our research, product, and organizational genes based on efficiency as the first principle!
Walking one step ahead of others and fighting battles where the few defeat the many – that’s the resounding Little Canon spirit!
Looking back over the past year, we did several things right:
- Advanced cognition, predicting technology and product directions with smaller samples. In the second half of 2024, the industry suddenly reached a new consensus that smaller models with higher knowledge density and edge-side intelligence mark a new stage in large model development. Now, due to the profound influence of edge-side intelligence development, mainstream consumer electronics and emerging hardware are evolving into super intelligent agents executing specific tasks in different scenarios, becoming a new tech entrepreneurship boom. Mianbi Intelligence discovered the huge research space and application potential inherent in edge-side intelligence more than six months ahead of the global curve. In this rapidly changing, competitive, and imaginative market, we’ve successfully built our foundation by heavily investing in edge-side models and creating the globally acclaimed Mianbi “Little Canon” MiniCPM.
- Technical foundation building, pragmatically serving vertical industries with advanced technology, rapid reuse, and value amplification. High-quality data from one industry can quickly empower that industry to build foundation models. Leveraging our technical background in text foundation models and deep accumulation of professional data, in July this year, Mianbi Intelligence, People’s Court Press, and Shenzhen Dibo jointly helped Shenzhen Intermediate Court officially launch the nation’s first judicial trial vertical domain large model, with applications covering all common civil and commercial cases and administrative cases. In November, the Faxin Legal Foundation Large Model, which Mianbi Intelligence participated in as a joint development team, was released by the Supreme Court. The Faxin Legal Foundation Model is positioned as a national-level legal AI infrastructure. In the future, thousands of courts, procuratorates, and judicial system units across the country will be places where we can make contributions.
- Achieving more with less, proposing Mianbi’s Law and Density Law for large models, scientifically surpassing similar products through large model development. Since the launch of Mianbi “Little Canon” MiniCPM edge-side model series in February this year, cumulative downloads have exceeded 3 million, frequently topping global renowned open-source community GitHub and HuggingFace’s large model trend rankings. MiniCPM not only consistently leads ahead in multiple benchmark tests but also pioneered the integration of unlimited long text, ultra-HD OCR recognition, and real-time video understanding into edge-side applications, creating multiple records. Behind Mianbi Little Canon’s ability to achieve more with less and high efficiency at low cost is our insistence on scientific large model development, guided by the more essential model knowledge density (knowledge density = model capability / model parameters involved in computation) in technology development and model iteration. We continue to innovatively propose leading large model scientific methodologies such as Mianbi’s Law and Density Law, aspiring to bring higher-performance edge-side intelligent technology to thousands of households and hundreds of industries.
All leadership ultimately comes from cognitive leadership; any business competition ultimately comes down to efficiency competition. Efficiency is not only the core of our business operations; in the business environment, it also means faster response times, lower costs, higher output, and better customer satisfaction. Thank you to all colleagues at Mianbi Intelligence, we are creating a miracle of extremely efficient large model development and business! Resilient, tenacious, and full of energy, Mianbi Little Canon has begun to shake the world. The Little Canon spirit is leading us toward the final victory of AGI!