Where is 面壁智能 headquartered?

面壁智能 is headquartered in Beijing, China. The company has a team of 101-500 people.

What does 面壁智能 do?

面壁智能 operates in the Foundation Models / LLMs sector. A Beijing-based AI company specializing in efficient edge-side (端侧) large language models, known for the MiniCPM series that delivers GPT-4-level performance on consumer devices. Efficient, lightweight edge-side LLMs that deliver GPT-4-class performance on consumer devices (phones, PCs, cars, wearables) with minimal compute and memory footprint — "以小博大" (small beating big).

When was 面壁智能 founded?

面壁智能 was founded in 2022. It is based in Beijing, China.

Who is the CEO of 面壁智能?

李大海 (Li Dahai) is the CEO of 面壁智能. The company was founded in 2022.

Where is 面壁智能 located?

面壁智能 is headquartered in Beijing, China.

What funding stage is 面壁智能 at?

面壁智能 is currently at the Growth Stage stage.

How many employees does 面壁智能 have?

面壁智能 has approximately 101-500 employees.

面壁智能

Category: Foundation Models / LLMs

A Beijing-based AI company specializing in efficient edge-side (端侧) large language models, known for the MiniCPM series that delivers GPT-4-level performance on consumer devices. 面壁智能 was founded in 2022. The company is led by 李大海 (Li Dahai). Based in Beijing, China. Team size: 101-500. Total funding raised: $358.4M. Latest round: Series C. Key investors include 知乎 (Zhihu), 华为哈勃 (Huawei Hubble), 春华创投 (Primavera), 中国电信 (China Telecom), 深创投 (Shenzhen Capital), 汇川产投 (Inovance), 茅台基金 (Moutai Fund), 洪泰基金 (Hongtai), 国中资本 (Guozhong), 龙芯创投 (Loongson), 鼎晖百孚 (CDH), 中关村科学城基金.

Founded: 2022
Headquarters: Beijing, China
Team size: 101-500
Total funding: $358.4M

Value proposition

Efficient, lightweight edge-side LLMs that deliver GPT-4-class performance on consumer devices (phones, PCs, cars, wearables) with minimal compute and memory footprint — "以小博大" (small beating big).

Products and solutions

MiniCPM series (edge-side LLMs: MiniCPM 4B/2.4B/1.2B/8B), MiniCPM-V (multimodal vision-language models), MiniCPM-o (omni-modal with voice), CPM-Bee (base model), cpmGO (edge-side AI assistant), AI Box for automotive (with Intel, 180 TOPS), 法信法律基座大模型 (legal AI), ChatDev (AI software development agent)

Unique value

Pioneer of the "Knowledge Density Law" (面壁定律) — achieving exponentially higher model capability per parameter over time. MiniCPM models achieve GPT-4V-level multimodal performance at 4B-8B parameters, running entirely on-device without cloud dependency.

Target customer

OEMs in automotive (smart cockpit), smartphone manufacturers, PC makers, robotics companies, legal institutions, and consumer electronics brands seeking on-device AI capabilities.

Industries served

Automotive (smart cockpits), Legal (court AI systems), Consumer Electronics (phones, PCs, wearables), Robotics (embodied AI), Education

Technology advantage

Proprietary "Knowledge Density Law" framework; MiniCPM models with extreme parameter efficiency (4B outperforming 13B+ models); Sparse attention mechanisms (InfLLM v2); CPM.cu custom CUDA inference framework; 220x inference speedup; Native support for CPU/GPU/NPU heterogeneous computing; Open-source ecosystem (OpenBMB community, Apache-2.0 licensed models)

How they differentiate

Unlike competitors pursuing ever-larger models, 面壁智能 focuses on extreme efficiency — achieving GPT-4-class performance at 4B-8B parameters that run on consumer devices. They coined the "Knowledge Density Law" showing model capability per parameter doubles every 3.3 months. Their models are fully open-source (Apache-2.0) and optimized for Intel, Qualcomm, MediaTek, and Huawei chips.

Main competitors

DeepSeek, 智谱AI (Zhipu AI), 月之暗面 (Moonshot AI / Kimi), 百川智能 (Baichuan Intelligent), 零一万物 (01.AI), Mistral AI (international edge-side competitor)

Key partnerships

Intel (strategic partner for AI Box & smart cockpit), Qualcomm (edge model optimization), MediaTek (chip adaptation), Baidu Cloud (端云协同), Huawei (strategic partner), Lenovo, Geely Auto (吉利银河 M9搭载), Changan Mazda (EZ-60搭载), Great Wall Motors, Changan Auto, SAIC Volkswagen, Booster Robotics (embodied AI)

Notable customers

China Supreme People's Court (法信法律基座大模型), Geely Auto (吉利银河 M9), Changan Mazda (EZ-60), Great Wall Motors, Changan Auto, SAIC Volkswagen, Huawei (端侧合作), multiple global smartphone OEMs (under NDA)

Major milestones

2022.08: Company founded, 2023.04: Angel round led by Zhihu, 2024.02: MiniCPM 2.4B released, outperforming Mistral-7B, 2024.04: Series A with Huawei Hubble, 2024.06: Stanford team caught plagiarizing MiniCPM ("被美国人抄袭"), 2024.07: MiniCPM-V published in Nature Communications, 2025.04: Strategic partnership with Intel for smart cockpit, 2025.05: Series B+ with Moutai Fund, 2026.02: Series C led by China Telecom, 2026.04: Unicorn valuation, Series C+ with Shenzhen Capital, 2026.04: World's first mass-produced AI Box for automotive (180 TOPS, with Intel)

Growth metrics

MiniCPM models exceeded 24M+ downloads across GitHub, HuggingFace, ModelScope platforms. Achieved unicorn valuation (>$1B) in 2026. '南Qwen,北面壁' dual open-source ecosystem leader. Team of 100+ with 80% from Tsinghua/Peking University.

Market positioning

Leading Chinese edge-side AI model company, positioned as the "端侧大模型先行者" (edge-side LLM pioneer). Ranked among China's top AI foundation model unicorns alongside Zhipu AI, Moonshot AI, Baichuan, and 01.AI. Recognized by Fortune as one of the top 50 AI innovators globally.

Geographic focus

China (primary); expanding globally through open-source community (HuggingFace, GitHub) and Intel partnership

Patents and IP

Multiple patents in efficient LLM training, sparse attention mechanisms (InfLLM v2), model quantization (BitCPM), and edge inference frameworks (CPM.cu). Models open-sourced under Apache-2.0 license.

About 李大海 (Li Dahai)

Ex-Google China (founding employee, 2006-2010); Ex-Yunyunwang (Engineering Director); Ex-Wandoujia (Search Tech Lead); Ex-Zhihu (CTO, Partner, 2015-2024). Peking University Master's in Mathematics.

Latest news about 面壁智能

More Foundation Models / LLMs companies

Official website: https://modelbest.cn