Skip to main content

ModelBest

Category: Foundation Models / LLMs

A pioneer in efficient, on-device Artificial General Intelligence (AGI) that develops lightweight, high-performance foundation models for edge computing and mobile hardware. ModelBest was founded in 2022. The company is led by Li Dahai. Based in Beijing, China. Team size: 100-200. Total funding raised: $100.0M. Latest round: Series B+ / Strategic Round ($40.0M est., Dec 2024). Key investors include ["Primavera Capital","Habo (Huawei)","Zhihu","Loongson Technology","CDH Investments","IDG Capital"].

Founded
2022
Headquarters
Beijing, China
Team size
100-200
Total funding
$100.0M

Value proposition

Delivers 'GPT-4 level' multimodal capabilities on consumer devices (phones/cars) with significantly lower latency, cost, and energy consumption, while ensuring data privacy by processing information locally.

Products and solutions

["MiniCPM Series (Flagship on-device LLMs, e.g., MiniCPM 3.0, MiniCPM-V)","MiniCPM-V (Multimodal model for image/video understanding on edge devices)","Luca (Consumer-facing AI native assistant/agent)","OpenBMB (Open-source community and model library)","CPM-Bee (Base foundation model for enterprise)"]

Unique value

Focuses on the 'Golden Ratio' of model scaling—achieving maximum intelligence with minimum parameter size (e.g., 2B parameters) to enable ubiquitous AI deployment.

Target customer

Smartphone OEMs, Automotive Manufacturers (Smart Cockpits), IoT/Robotics companies, and Enterprise developers requiring private/local AI.

Industries served

["Consumer Electronics (AI Smartphones/PCs)","Automotive (Intelligent Cockpits)","Robotics (Embodied AI)","Internet Services","Education"]

Technology advantage

Possesses a 'Full-Stack' optimization capability ranging from model architecture (proprietary training methods) to hardware adaptation (NPU optimization), validated by deep integration with chipmakers like Qualcomm and MediaTek.

How they differentiate

Specializes in 'efficient, on-device' foundation models (MiniCPM series) rather than just massive cloud-based models. Focuses on the 'Golden Ratio' of performance-to-size (e.g., 2B-8B parameters) to enable LLMs on smartphones, cars, and IoT devices with lower latency and cost.

Main competitors

["DeepSeek","Zhipu AI","MiniMax","01.AI (Yi)"]

Key partnerships

["Hardware/Chipset: Huawei (Habo), MediaTek, Qualcomm, Intel","Automotive: Volkswagen, Geely, Changan Automobile, Great Wall Motor","Strategic: Zhihu (Data & Investment), Tsinghua University (R&D)"]

Notable customers

["Volkswagen","Geely","Changan Automobile","Great Wall Motor","Zhihu"]

Major milestones

["Founded in Aug 2022 as a spin-off from Tsinghua University NLP Lab","Secured Angel investment from Zhihu in April 2023","Released MiniCPM 'Little Steel Cannon' series in early 2024","Completed Series B financing led by Huawei's Habo and Primavera in April 2024","Closed Series B+ round with Loongson Technology in Dec 2024"]

Growth metrics

MiniCPM models deployed at scale on devices from Huawei, Qualcomm, and MediaTek; ChatDev open-source project garnered 12k+ stars on GitHub; Partnered with major auto OEMs for smart cockpit AI.

Market positioning

Leader in Edge AI and On-Device LLMs; Challenger to generalist model providers (like Baidu/Alibaba) in the mobile/IoT sector.

Geographic focus

China (Primary), Global (via Open Source communities like Hugging Face/GitHub)

Patents and IP

Intellectual Property derived from Tsinghua University NLP Lab transfer; specific patent portfolio numbers are undisclosed but backed by extensive academic citations.

About Li Dahai

Former CTO and Senior Vice President of Zhihu; Founding employee of Google China; Engineering Director at Yunyun.com; Head of Search Technology at Wandoujia.

Official website: