# 闫俊杰 · Yan Junjie — 封面传记 ACF-00-00133

> 亚洲封面人物 Asia Cover Figure · 机器可读档案（LLM-ready）。本文件由官网结构化档案数据自动生成，供 AI 检索与引用。中文全文与英文全文对照编排。

## 档案元数据 Metadata

- 封面编码 ACF Code：**ACF-00-00133**
- 姓名 Name：闫俊杰 / Yan Junjie
- 职务 Title：创始人、CEO兼CTO @ MiniMax（稀宇科技） / Founder, CEO & CTO @ MiniMax (Shanghai Xiyu Technology)
- 公司 Company：稀宇科技（上海）有限公司 / MiniMax Group Inc.
- 篇别 Category：格局（格局篇 / Cover Biography (Geju)）
- 入档日期 Accessioned：2026-02-13
- 标签 Tags：人工智能, 大模型, AGI, MoE, 开源, 全球化, AI六小虎, 科创板
- 永久档案链接 Archive URL：https://coverfigure.com/acf/ACF-00-00133/geju
- English archive：https://coverfigure.com/acf/ACF-00-00133/geju?lang=en
- 官网原文报道 Feature story：https://coverfigure.com/acf/figure/yanjunjie

## 信息口径声明 Editorial Sourcing Policy

- 工商登记信息可通过国家企业信用信息公示系统核验；经营数据来自企业公开披露、官方发布与权威媒体报道，引用处标注来源。
- 收购、IPO、估值、融资等事项以官方公告为准；人物观点与言论为公开演讲、采访的事实记录，不代表编辑部立场。
- 本档案仅供行业研究参考，不构成投资建议。事实更正联系：editorial@coverfigure.com。

## 卷首 Editorial Intro

四年。七轮融资。十五亿美元。从零到IPO。这组数字在全球AI产业的时间轴上刻下了一道几乎不可思议的速率标记。从公司成立到登陆资本市场，MiniMax刷新了全球AI企业从创立到上市的最短时间纪录，与大洋彼岸的OpenAI、Anthropic站在了同一张时间表的对照栏里。而这家公司背后的核心人物，在创业启动时只有三十三岁，办公地点是上海一间不到一百平方米的房间，初始团队人数个位数。

这个人叫闫俊杰。在此之前，他是商汤科技最年轻的副总裁，负责过深度学习框架的核心研发。但真正让他在全球AI版图中留下独特印记的，不是他在大公司的晋升速度，而是他离开大公司后做出的一系列"非共识"选择。当行业共识指向稠密模型时，他押注混合专家架构；当多数公司追求闭源壁垒以保护竞争优势时，他率先走向开源；当大多数中国AI公司聚焦国内市场时，他从第一天就锁定了全球化。

这些选择中的每一个，在当时都被相当比例的业内人士视为冒险甚至冒进。但四年后的数据给出了清晰的验证——MiniMax的混合专家架构用更低的算力成本实现了可比肩顶级稠密模型的智能水平；开源策略为其赢得了全球开发者社区的广泛响应；Talkie产品在全球市场获得数千万用户，证明了AI应用出海的可行路径。

这些数字背后是一个更深层的叙事：中国AI产业从"跟跑"到"并跑"再到局部"领跑"的结构性转折。闫俊杰用博弈论中的"极小化极大"算法哲学指导创业——在最坏条件下寻找最优解，用最小化成本最大化智能。亚洲《封面人物》杂志独立编委会决定将闫俊杰的档案纳入永久典藏。终身档案编码记录这位从河南县城走出的年轻技术领袖如何重新定义中国AI创业的可能性。

In early 2022, while the global AI industry was still awaiting its ignition point on the eve of ChatGPT, in an office of less than 100 square meters in Shanghai, a thirty-three-year-old young man and several colleagues wrote down the company's founding vision and path: 'Intelligence with Everyone.' This person was Yan Junjie, previously SenseTime's youngest vice president. His company was called MiniMax, named after the minimax algorithm in game theory, finding the optimal solution under the worst conditions. This naming almost prophesied the entrepreneurial script of the next four years: leveraging minimal resources to maximize intelligence in an environment of limited computing power, fierce competition, and highly divergent technical routes. Three years later, MiniMax refreshed the global record for the fastest AI company from founding to IPO in approximately three years, becoming one of the 'North-South Duo' alongside Zhipu. This is not merely an entrepreneurial story but a microcosm of the structural turning point in China's AI industry from following to paralleling to leading in select areas.

## 人物速览 Lead

从河南县城到全球AI一代标杆，闫俊杰以非共识的技术路线和极致的成本控制，四年将MiniMax打造为从创立到IPO最快的AI公司。他押注MoE架构、坚持全模态自研、率先开源，用“最小化成本、最大化智能”的哲学重新定义了中国AI创业的可能性。

From a Henan county town to a global AI generation benchmark, Yan Junjie built MiniMax into the fastest AI company from founding to IPO in four years through contrarian technical routes and extreme cost control. By betting on MoE architecture, insisting on full-modal self-development, and pioneering open-source releases, he redefined the possibilities of Chinese AI entrepreneurship with the philosophy of 'minimizing cost, maximizing intelligence.'

## 正文 Archive Chapters

### 1. 县城少年：底层逻辑的被动养成 / County Town Youth: The Passive Cultivation of Foundational Logic

1989年，闫俊杰出生在河南商丘夏邑县的一座小县城。这是一个典型的中原内陆县城，经济以农业和低端制造业为主，与“科技创新”没有任何可见的关联。

闫俊杰的青少年时代，用他自己的话来说：“很多东西没人教，只能靠自己摸索。”这句话看似平淡，却蕴含了他日后创业精神的底层密码。在一个缺乏信息、缺乏引导、缺乏资源的环境中成长，迫使闫俊杰被动地培养了一种能力——面对从未做过的事情时，能迅速找到一些底层逻辑。

2006年，闫俊杰考入东南大学数学系。选择数学作为本科专业，显示了他对抽象思维和逻辑推理的天然偏好。数学训练为他日后在人工智能领域的研究奠定了坚实的数理基础——无论是后来在混合专家系统（MoE）架构中对稀疏性的精确把控，还是在博弈论思维中寻找最优解的商业直觉，都可以追溯到东南大学数学系的四年训练。

本科毕业后，闫俊杰进入中国科学院自动化研究所硕博连读，研究方向是模式识别与人工智能。彼时大约是2010年前后，深度学习尚未在全球范围内引爆，但计算机视觉和模式识别领域已经暗流涌动。闫俊杰在中科院的研究方向——人脸识别和视频物体检测——正是后来商汤科技的核心技术赛道，也是他职业生涯的第一块基石。

博士期间，闫俊杰以实习生身份加入百度研究院。虽然实习时间不长，但这段经历让他打破了“学术与应用脱节”的桎梏，第一次感受到了AI技术的产业价值。用他自己的话说，这是从“论文”到“产品”的认知跨越。

2014年11月，在百度实习结束后，闫俊杰加入了初创不久的商汤科技。这个决定在当时看来颇为冒险——商汤还只是一家刚成立的小公司，远未达到后来的“AI四小龙”地位。但闫俊杰的判断是准确的：在一个技术驱动的新兴领域，加入一家处于上升期的创业公司，比去大厂做螺丝钉有更大的成长空间。这个选择本身就体现了他的“极小化极大”思维——在有限信息条件下做出最优决策。

2015年7月，闫俊杰获得中国科学院自动化研究所人工智能领域博士学位，随后在清华大学计算机系从事博士后研究。学术训练的完成，为他日后在商汤和MiniMax的技术领导角色提供了坚实的信用背书。

**English:** In 1989, Yan Junjie was born in a small county town in Xiayi County, Shangqiu, Henan Province. This is a typical inland Central Plains county, with an economy based on agriculture and low-end manufacturing, with no visible connection to 'scientific and technological innovation.'

Yan Junjie's youth, in his own words: 'Many things no one taught me; I could only explore on my own.' This seemingly mundane statement contains the underlying code of his future entrepreneurial spirit. Growing up in an environment lacking information, guidance, and resources forced Yan Junjie to passively cultivate an ability: when facing something he had never done before, he could quickly find some foundational logic.

In 2006, Yan Junjie was admitted to the Department of Mathematics at Southeast University. Choosing mathematics as his undergraduate major revealed his natural preference for abstract thinking and logical reasoning. Mathematical training laid a solid mathematical foundation for his future research in AI, whether it was the precise control of sparsity in the Mixture of Experts (MoE) architecture or the business intuition of finding optimal solutions through game theory thinking, both traceable to four years of training in Southeast University's mathematics department.

After his bachelor's degree, Yan Junjie entered a combined master's-doctoral program at the Institute of Automation, Chinese Academy of Sciences, researching pattern recognition and artificial intelligence. Around 2010, deep learning had not yet exploded globally, but computer vision and pattern recognition were already surging beneath the surface. Yan Junjie's research direction at CAS, facial recognition and video object detection, was precisely the core technology track that SenseTime would later pursue and the first cornerstone of his career.

During his doctoral studies, Yan Junjie joined Baidu Research as an intern. Although the internship was brief, this experience broke the shackles of 'academic-application disconnect' and gave him his first taste of AI technology's industrial value. In his own words, this was the cognitive leap from 'papers' to 'products.'

In November 2014, after completing his Baidu internship, Yan Junjie joined the newly founded SenseTime. This decision appeared quite risky at the time, as SenseTime was still a small startup far from its later 'AI Four Dragon' status. But Yan Junjie's judgment proved accurate: in a technology-driven emerging field, joining a startup on the rise offered greater growth potential than being a cog in a big company. This choice itself reflected his 'minimax' thinking, making optimal decisions under limited information.

In July 2015, Yan Junjie earned his PhD in artificial intelligence from the Institute of Automation, CAS, and subsequently conducted postdoctoral research in the Department of Computer Science at Tsinghua University. The completion of his academic training provided solid credibility for his future technical leadership roles at SenseTime and MiniMax.

### 2. 商汤七年：从实习生到最年轻副总裁 / Seven Years at SenseTime: From Intern to Youngest VP

2014年底加入商汤科技后，闫俊杰在短短七年内完成了从研究员到集团副总裁的跨越。

他在商汤内部创立了“目标感知组与人脸识别组”，大幅提高了相关技术的准确率和应用水平。2017年，闫俊杰升任商汤科技研究院副院长，开始参与公司的核心技术战略制定。2019年1月，他被任命为智慧城市事业群CTO兼研究院副院长，带领一个超过七百人的跨职能团队，将面部识别算法准确率提升至行业第一，落地安防、智慧城市等核心场景。

在商汤的七年，是闫俊杰从“学者”蜕变为“技术企业家”的关键期。他不仅积累了大规模团队管理和商业化落地的经验，更亲历了中国AI产业从1.0时代（计算机视觉）到2.0时代（大模型）的完整周期。一位AI四小龙创始人曾如此评价他：“他吃过AI 1.0的苦。”

这句话的含义是：闫俊杰完整经历了上一轮AI热潮的起伏。他曾在商汤做人脸识别时经历过“一年半一直输”的低谷，顶着巨大压力逆风追赶后才迎来突破。这种从逆境中锻造出的技术自信和韧性，成为他日后创业的精神支柱。

2019年4月，一件看似与闫俊杰工作无关的事情深刻改变了他的人生轨迹。美国OpenAI公司的AI战队在DOTA2游戏中击败了2018年冠军OG战队，这是AI首次在电子竞技中击败人类世界冠军。作为DOTA2忠实玩家的闫俊杰马上钻研了OpenAI发布的相关论文，并意识到了一个关键问题：当时的AI虽然厉害，但每次完成特定任务都需要专门训练一个模型，费时费力且只服务于少数大客户。

如果能开发一个通用模型，不用在每次完成任务前专门训练，就能为普通用户服务了。这个洞察成为日后MiniMax创业的核心逻辑。

2021年底，商汤科技即将在港股上市。彼时的闫俊杰已经是商汤最年轻的副总裁、研究院副院长。但他做出了一个让很多人意外的决定：离开商汤，创业。他曾在一次采访中说：“两个同样优秀的毕业生，一个去了大厂，一个去了创业公司。两三年后，选择创业的人成才率更高。”

**English:** After joining SenseTime in late 2014, Yan Junjie completed the leap from researcher to group vice president in just seven years.

He founded the 'Object Perception Group and Face Recognition Group' within SenseTime, significantly improving the accuracy and application level of related technologies. In 2017, Yan Junjie was promoted to Deputy Director of SenseTime's Research Institute, beginning to participate in the company's core technology strategy formulation. In January 2019, he was appointed CTO of the Smart City Business Group and Deputy Director of the Research Institute, leading a cross-functional team of over 700 people, elevating facial recognition algorithm accuracy to industry-first levels and deploying core scenarios in security and smart cities.

The seven years at SenseTime were the critical period for Yan Junjie's transformation from 'scholar' to 'technology entrepreneur.' He not only accumulated experience in large-scale team management and commercial deployment but also witnessed the complete cycle of China's AI industry from the 1.0 era (computer vision) to the 2.0 era (large models). One AI Four Dragon founder evaluated him thus: 'He has tasted the bitterness of AI 1.0.'

The meaning behind this statement is that Yan Junjie fully experienced the rise and fall of the previous AI wave. During his time working on facial recognition at SenseTime, he went through a low period of 'losing for a year and a half,' enduring enormous pressure while catching up against the wind before achieving a breakthrough. This technical confidence and resilience forged from adversity became the spiritual pillar for his future entrepreneurship.

In April 2019, something seemingly unrelated to Yan Junjie's work profoundly changed his life trajectory. OpenAI's AI team defeated the 2018 DOTA2 champion OG team in a game, the first time AI defeated human world champions in competitive e-sports. As a devoted DOTA2 player, Yan Junjie immediately studied OpenAI's related papers and identified a critical problem: while AI was impressive, each specific task required training a dedicated model, which was time-consuming, labor-intensive, and served only a few large clients.

What if a general-purpose model could be developed, eliminating the need for task-specific training, thereby serving ordinary users? This insight became the core logic of the future MiniMax venture.

By late 2021, SenseTime was about to list on the Hong Kong Stock Exchange. Yan Junjie was already SenseTime's youngest vice president and deputy research institute director. But he made a decision that surprised many: leave SenseTime and start a company. He once said in an interview: 'Two equally excellent graduates, one joins a big company, one joins a startup. Two or three years later, the one who chose the startup has a higher success rate.'

### 3. 创业起点：一百平方米里的五年规划 / The Starting Point: Five-Year Plan in 100 Square Meters

2021年12月9日，MiniMax组建团队的第一天，闫俊杰和几个同伴花了九个小时写下五年规划。他们在一间简陋的办公室里，在白板上拆解技术变化和市场机会，定下了三个原则：一定要直接服务用户、一定要做国际化、一定要技术驱动。

2022年初，闫俊杰从商汤科技正式离职，在上海创立MiniMax。公司最初的办公室只有100多平方米，“搬了很多次家，最后落定到上海市徐汇区”，只有几十个人。MiniMax这个名字，源自博弈论中的minimax算法——在最坏的条件下找到最优解。公司愿景则被定义为“Intelligence with Everyone”（与所有人共创智能）。

闫俊杰在创业第一天就坚定押注AGI（通用人工智能）。这在当时是一个“非共识”判断——2022年初，AGI在中国还是小众概念，绝大多数人甚至不知道这个词意味着什么。但闫俊杰的洞察来自对技术趋势的深刻理解：旧式AI模型因不够通用，需为不同场景重复训练；而大模型秉持“One Model for all”理念，有望突破传统定制化to B/G的局限。

联合创始人贠烨祎的加入为MiniMax补齐了关键拼图。贠烨祎毕业于约翰斯-霍普金斯大学，曾在商汤担任CEO办公室战略负责人和融资与战略投资部经理。她与闫俊杰形成了绝佳的互补：一个专注定义技术愿景与产品未来，一个主导战略执行、资本对接与复杂沟通。

凭借在商汤积累的人脉，贠烨祎为MiniMax引入了两位重要的早期支持者：云启资本管理合伙人陈昱和IDG合伙人牛奎光。2022年初，MiniMax完成了天使轮融资——米哈游、IDG资本和高瓴资本共同注资3100万美元，投前估值1.7亿美元。高瓴是第一个投资人，合伙人在与闫俊杰聊了三小时后，拿出了一张估值为空白的投资意向书：“你们可以写一个想要的估值和投资金额。”闫俊杰给出了计划中的数字：融资3000万美元，估值2亿美元。团队坚持了自身节奏，并未接受更高估值。

高瓴的判断基于一个核心观察：闫俊杰在商汤带过1000多人的团队，在算法、工程、组织和AI商业化上都有经验，这是一个罕见的“全能型”AI创始人。不到两周，高瓴走完了投资决策流程。

**English:** On December 9, 2021, the first day MiniMax assembled its team, Yan Junjie and several colleagues spent nine hours writing a five-year plan. In a modest office, they deconstructed technological changes and market opportunities on a whiteboard, establishing three principles: must directly serve users, must go international, must be technology-driven.

In early 2022, Yan Junjie officially left SenseTime and founded MiniMax in Shanghai. The company's initial office was just over 100 square meters, 'moving many times before finally settling in Xuhui District, Shanghai,' with only dozens of people. The name MiniMax originated from the minimax algorithm in game theory, finding optimal solutions under worst conditions. The company's vision was defined as 'Intelligence with Everyone.'

From the very first day, Yan Junjie firmly bet on AGI (Artificial General Intelligence). This was a 'contrarian' judgment at the time: in early 2022, AGI was still a niche concept in China, with most people not even knowing what the term meant. But Yan Junjie's insight came from deep understanding of technology trends: old-style AI models, being insufficiently general, needed to be retrained for different scenarios; whereas large models embracing the 'One Model for all' philosophy could break through the limitations of traditional customized to B/G approaches.

Co-founder Yun Yi's joining completed a key puzzle piece for MiniMax. Yun graduated from Johns Hopkins University and had served as head of strategy in the CEO's office and manager of financing and strategic investment at SenseTime. She formed an excellent complement with Yan Junjie: one focused on defining technical vision and product future, the other leading strategic execution, capital interfacing, and complex communications.

Leveraging connections from SenseTime, Yun Yi introduced two important early supporters to MiniMax: Yunqi Capital managing partner Chen Yu and IDG partner Niu Kuiguang. In early 2022, MiniMax completed its angel round: Mihayou, IDG Capital, and Hillhouse Capital co-invested 31 million USD at a pre-money valuation of 170 million USD. Hillhouse was the first investor; after talking with Yan Junjie for three hours, partners presented a term sheet with a blank valuation: 'You can write whatever valuation and investment amount you want.' Yan Junjie gave the planned figures: 30 million USD funding at 200 million USD valuation. The team maintained its own rhythm and did not accept higher valuations.

Hillhouse's judgment was based on a core observation: Yan Junjie had led teams of over 1,000 at SenseTime, with experience spanning algorithms, engineering, organization, and AI commercialization, making him a rare 'all-round' AI founder. In less than two weeks, Hillhouse completed its investment decision.

### 4. 非共识之路：押注MoE与全模态自研 / The Contrarian Path: Betting on MoE and Full-Modal Self-Development

2023年下半年，当同行大多还在迭代更稳妥的“稠密模型”时，闫俊杰做出了一个在当时看来极具争议的决定：将公司绝大部分研发与算力资源投入非主流的“混合专家系统”（MoE）路线。

MoE路线的核心逻辑是：通过稀疏激活机制，在推理时只激活部分专家网络，从而以更低的训练和推理成本达到接近甚至超越稠密模型的效果。但在当时，这条路充满不确定性——行业内很少有人成功走通过。

研发之路异常艰辛。MiniMax经历了三到四次严重的预训练失败，每一次模型崩溃就意味着两个月的研发周期白费，以及单次高达1500万美元的算力成本打了水漂。但闫俊杰没有退缩。他认为：“只有把MoE做出来，才能用更低的成本向更高的模型阶段演进，才能追赶上OpenAI。”

投资人黄明明后来感叹：闫俊杰外表温和，实则是个“狠人”，“在资源有限的情况下，他把当时手头80%的算力资源都用来推进MoE。只有这样的企业家才能走得更远，走到终局。”

与此同时，闫俊杰从一开始就坚持全模态自研路线——同时布局文本、语音和视频模型的自主研发。很多人建议他聚焦一个模态，但闫俊杰的判断是：“真正的AGI一定是不分模态的。”团队的逻辑是：AGI必须是多模态的输入和输出，哪怕再难，每个模态也要先独立走通。

这种非共识路线在2024年迎来了验证。2024年4月，MiniMax发布了创新的MoE模型，在多项基准测试中达到国际顶尖水平。闫俊杰提出的混合专家稀疏度在2%以内仍能保持良好收敛性，这一发现令行业瞩目。同时，他提出的Sparse Attention架构在保持训练稳定性的同时大幅压缩了长上下文的计算效率。

2025年6月，MiniMax-M1模型开源发布，以较低的算力成本显著提升了深度推理效率，实现了对OpenAI的弯道超车。当时，闫俊杰在朋友圈写道：“第一次感觉到大山不是不能翻越。”

2026年8月的中期业绩会上，闫俊杰进一步阐释了公司的技术路线：“Minimize the Inference Cost，Maximize the Intelligence”实现“Intelligence with Everyone”。他将其定义为“不只是提供有性价比的服务，更是面向更高级智能进行有效Scaling的核心能力”。

**English:** In the second half of 2023, when most peers were still iterating on the safer 'dense model' approach, Yan Junjie made a highly controversial decision at the time: investing the vast majority of the company's R&D and computing resources into the non-mainstream 'Mixture of Experts' (MoE) route.

The core logic of MoE is: through sparse activation mechanisms, only activating part of the expert network during inference, achieving performance approaching or even exceeding dense models at lower training and inference costs. But at that time, this path was full of uncertainty: very few in the industry had successfully navigated it.

The R&D journey was extraordinarily grueling. MiniMax experienced three to four serious pre-training failures, each model collapse meaning two months of R&D cycles wasted and a single computing cost of up to 15 million USD lost. But Yan Junjie did not retreat. He believed: 'Only by making MoE work can we evolve to higher model stages at lower cost and catch up with OpenAI.'

Investor Huang Mingming later marveled: Yan Junjie appears gentle on the surface but is actually a 'ruthless person.' 'With limited resources, he committed 80% of his available computing power to advancing MoE. Only such an entrepreneur can go further and reach the endgame.'

Meanwhile, Yan Junjie insisted on a full-modal self-development route from the beginning, simultaneously laying out independent R&D for text, speech, and video models. Many advised him to focus on one modality, but Yan Junjie's judgment was: 'True AGI is certainly modality-agnostic.' The team's logic was: AGI must involve multi-modal input and output; no matter how difficult, each modality must first independently work.

This contrarian route was vindicated in 2024. In April 2024, MiniMax released its innovative MoE model, achieving international top-tier levels in multiple benchmark tests. Yan Junjie's discovery that hybrid expert sparsity below 2% could still maintain good convergence attracted industry attention. Simultaneously, his proposed Sparse Attention architecture greatly compressed long-context computational efficiency while maintaining training stability.

In June 2025, MiniMax-M1 model was open-sourced, significantly improving deep reasoning efficiency at lower computing costs, achieving a catch-up overtake against OpenAI. At the time, Yan Junjie wrote on his social media: 'For the first time, I feel the mountain is not impossible to cross.'

At the August 2026 interim results meeting, Yan Junjie further elaborated the company's technical route: 'Minimize the Inference Cost, Maximize the Intelligence' to achieve 'Intelligence with Everyone.' He defined this as 'not just providing cost-effective services, but a core capability for effective Scaling toward higher intelligence.'

### 5. 产品化突围：从Glow到Talkie的全球征途 / Product Breakthrough: From Glow to Talkie's Global Campaign

MiniMax在2022年10月低调发布了首款产品Glow——一款虚拟聊天AI应用。未做大规模推广，便在两个月内自然积累了上百万二次元用户。对于一家正式运营不足一年的创业公司，这是一个充满希望的开端。

然而，2022年11月ChatGPT的横空出世，让Glow的声量迅速被全球性巨浪淹没。闫俊杰团队迅速调整策略，在高瓴的牵线下引入前今日头条用户产品负责人张前川。张前川的“字节基因”让MiniMax的产品化进入快车道。

2023年6月，AI社交应用Talkie（星野的海外版）发布后迅速走红，一度跻身美国Google Play下载总榜TOP6，累计下载量超过1400万次。Talkie的成功在于精细的用户体验设计：UI特意选了金色和黑色以显华贵，星星元素恰到好处地出现在各处，对话框会自动设置括号体以配合用户的“画外音”习惯。

与此同时，国内版“星野”和AI语音助手“海螺AI”也在持续迭代。海螺AI生成的动物跳水系列短视频在TikTok和Instagram上拥有最高7000万的播放量。产品覆盖全球超过200个国家和地区。

2025年财报显示，MiniMax全年总收入7903.8万美元，同比增长158.9%。其中AI原生产品（C端）收入5307.5万美元，开放平台及企业服务（B端）收入2596.3万美元，同比暴增197.8%。海外收入占比超过70%，新加坡市场收入占24%，美国市场占20%。

但闫俊杰并非满足于C端产品的成功。2024年9月，星野及海外版Talkie的产品负责人张前川淡出日常管理。此后，MiniMax开始从“互联网产品公司”向“AI平台型公司”转型。2026年3月的业绩会上，闫俊杰提出核心观点：MiniMax要从大模型公司进化为AI时代的平台型公司。在他的定义中，AI平台的价值公式是：智能密度乘以Token吞吐能力。

到2026年8月，收入结构已发生根本性变化：B端业务占比达到约80%，C端约20%，对比去年同期B端仅30%。企业客户和开发者数量突破200万，是2025年底的10倍。8月年度经常性收入（ARR）超过8亿美元。

**English:** In October 2022, MiniMax quietly launched its first product, Glow, a virtual chat AI application. Without large-scale promotion, it naturally accumulated over a million ACG users within two months. For a startup that had been formally operating for less than a year, this was a promising start.

However, the emergence of ChatGPT in November 2022 quickly overwhelmed Glow's visibility with a global tidal wave. Yan Junjie's team swiftly adjusted strategy, and through Hillhouse's connection, brought in former Toutiao user product head Zhang Qianchuan. Zhang's 'ByteDance genes' accelerated MiniMax's product development.

In June 2023, the AI social app Talkie (the overseas version of Xingye) quickly went viral after launch, once reaching the TOP 6 of the US Google Play overall download chart, with cumulative downloads exceeding 14 million. Talkie's success lay in meticulous user experience design: the UI deliberately chose gold and black for luxury, star elements appeared恰到好处 throughout, and the dialog box automatically set bracket-style formatting to accommodate users' 'voiceover' habits.

Meanwhile, the domestic version Xingye and AI voice assistant Hailuo AI continued iterating. Short videos generated by Hailuo AI's animal diving series achieved up to 70 million views on TikTok and Instagram. Products covered over 200 countries and regions globally.

The 2025 financial report showed MiniMax's full-year total revenue of 79.038 million USD, up 158.9% year-on-year. AI-native products (C-end) contributed 53.075 million USD, while open platform and enterprise services (B-end) contributed 25.963 million USD, surging 197.8% year-on-year. Overseas revenue exceeded 70%, with Singapore at 24% and the US at 20%.

But Yan Junjie was not content with C-end product success. In September 2024, Xingye and Talkie's product head Zhang Qianchuan stepped back from daily management. Subsequently, MiniMax began transitioning from an 'internet product company' to an 'AI platform company.' At the March 2026 results meeting, Yan Junjie proposed the core thesis: MiniMax must evolve from a large model company into a platform company for the AI era. In his definition, the AI platform value formula is: intelligence density multiplied by Token throughput capacity.

By August 2026, the revenue structure had fundamentally shifted: B-end business accounted for approximately 80%, C-end about 20%, compared to B-end at only 30% the previous year. Enterprise clients and developers exceeded 2 million, ten times the end-2025 figure. August annual recurring revenue (ARR) exceeded 800 million USD.

### 6. 资本加速度：4年7轮15亿美元的融资历程 / Capital Acceleration: 1.5 Billion USD in 7 Rounds Over 4 Years

MiniMax的融资速度是中国AI大模型行业最快的之一。从ChatGPT问世前完成两轮融资，到上市前累计七轮、30家机构投资约15亿美元，MiniMax完成了一场中国AI的第一场资本盛宴。

天使轮（2022年初）：高瓴领投，米哈游、IDG资本、云启资本参投，3100万美元，估值2亿美元。闫俊杰给高瓴团队讲了九小时的“技术课”，从Transformer架构的Scaling Laws到GPT-3的进展、DeepMind的强化学习、扩散模型和CLIP。高瓴投资人薛子钊后来回忆：“当时很少有人能把这些技术拼到一起。”

Pre-A轮（2022年中）：明势资本领投。创始合伙人黄明明第一次见到闫俊杰时，他正专注地在iPad上阅读论文。两个多小时的首次交流，闫俊杰从技术变革的源头讲起，明确回答“to C，不走定制化to B的老路”。

A轮及以后：腾讯投资超过2.5亿美元，估值超过12亿美元。阿里巴巴后来领投6亿美元。红杉中国在2023年7月首次投资，后续持续加码。明势资本参与了八轮融资中的六轮，成为最坚定的陪伴者。

截止上市前，4年多时间里MiniMax累计融资近15亿美元。IPO后，阿里巴巴持股12.52%为最大外部股东，米哈游5.87%，高瓴7.14%为最大财务投资人，红杉中国3.49%，腾讯2.37%。

创始人闫俊杰通过AlphaEXP Limited持有19.96%的B类普通股，因B类股1票顶A类股10票，实际掌控59.98%的投票权，牢牢掌握了公司控制权。

2026年1月9日，MiniMax以每股165港元发行价在港交所挂牌上市，上市首日收盘大涨109%，市值突破千亿港元。基石投资者阵容显赫：阿布扎比投资局6500万美元、阿里巴巴3000万美元、Aspex和博裕各3500万美元，基石轮合计融资3.5亿美元。闫俊杰在上市敲钟前分享：“我们初步探索了一条纯草根AI创业的路径。”

**English:** MiniMax's fundraising pace was among the fastest in China's AI large model industry. From completing two rounds before ChatGPT's emergence to accumulating seven rounds totaling approximately 1.5 billion USD from 30 institutional investors before IPO, MiniMax completed China's first AI capital feast.

Angel round (early 2022): Hillhouse led, with Mihayou, IDG Capital, and Yunqi Capital participating. 31 million USD at 200 million USD valuation. Yan Junjie gave the Hillhouse team a nine-hour 'technology lecture,' from Transformer architecture Scaling Laws to GPT-3 progress, DeepMind reinforcement learning, diffusion models, and CLIP. Hillhouse investor Xue Zizhao later recalled: 'At that time, very few people could piece these technologies together.'

Pre-A round (mid-2022): Mingshi Capital led. Founding partner Huang Mingming first met Yan Junjie, who was focused on reading papers on his iPad. Over two hours of initial discussion, Yan Junjie explained technological transformation from its origins, clearly stating 'to C, not the old customized to B path.'

Series A and beyond: Tencent invested over 250 million USD at over 1.2 billion USD valuation. Alibaba later led 600 million USD. Sequoia China first invested in July 2023 and continued adding. Mingshi Capital participated in six of eight total rounds, becoming the most steadfast companion.

Before IPO, over four years MiniMax cumulatively raised nearly 1.5 billion USD. Post-IPO, Alibaba held 12.52% as the largest external shareholder, Mihayou 5.87%, Hillhouse 7.14% as the largest financial investor, Sequoia China 3.49%, Tencent 2.37%.

Founder Yan Junjie, through AlphaEXP Limited, held 19.96% B-class ordinary shares. Since B-class shares carry 10 votes per share versus A-class 1 vote, he effectively controlled 59.98% voting rights, firmly maintaining company control.

On January 9, 2026, MiniMax listed on HKEX at an IPO price of 165 HKD per share, with the stock surging 109% on the first day, market cap exceeding 100 billion HKD. The cornerstone investor lineup was stellar: Abu Dhabi Investment Authority 65 million USD, Alibaba 30 million USD, Aspex and Boyu each 35 million USD, totaling 350 million USD. Before ringing the bell, Yan Junjie shared: 'We have initially explored a path for pure grassroots AI entrepreneurship.'

### 7. 开源哲学与全球化：智能平权的技术路线 / Open-Source Philosophy and Globalization: The Technical Route to Intelligence Equity

闫俊杰的技术哲学中，有一个核心信条：“所有模型一年后都会落后，藏私没有意义。”

2025年1月，MiniMax首次将模型开源，发布并开源了MiniMax-01系列模型。2025年6月，MiniMax-M1模型开源发布。2026年，H3模型开源三个多星期以来，下载量已超过2400万次，产生超过300个公开衍生模型，成为2026年全球下载量最高的模型。

开源的战略意义在于：降低开发者和企业的使用门槛，使模型更快进入云平台、开发者工具及企业工作流。闫俊杰的开源逻辑不是慈善，而是精明——开源带来生态，生态带来标准，标准带来长期定价权。

全球化是MiniMax从第一天就确定的战略方向。闫俊杰一开始就锁定全球市场，“这样有更广阔的空间”。目前，MiniMax产品覆盖200多个国家和地区，全球用户约3亿，企业客户超200万。海外收入占比超过70%，这在中国AI公司中极为罕见。

2026年6月1日，MiniMax正式发布新一代通用模型M3，具备更长上下文、更稳定记忆、更低成本推理能力。M3发布不到两周，华为云即宣布完成对M3的开源首发适配，基于昇腾算力底座提供算力支持。摩尔线程、沐曦股份等国产GPU厂商也已完成适配。

闫俊杰在中期业绩会上表示：“大语言模型带动的智能提升几乎没有尽头。”从Coding和Agent能力的实用，到更自主地完成长程任务，再到未来端到端交付可依赖的结果，MiniMax持续在追求更高的智能。

2026年1月，闫俊杰在《人民日报》撰文写道：“基于大模型的人工智能研发，投入巨大，每一代模型的研发周期往往长达半年到一年，因此研究方向的选择非常关键。有两件事很重要：一是模型的技术指标能够触达智能水平的最前沿；二是模型要有自己的特色，并且契合生产力变革的发展趋势。”

从更深层的战略逻辑来看，闫俊杰的全球化路线和开源策略构成了一个相互强化的飞轮。开源带来全球开发者社区的快速反馈和生态建设，全球化则为开源模型提供了最大化的应用场景和商业价值。这种“开源+全球化”的组合拳，在中国AI创业公司中几乎没有先例，却恰恰是MiniMax能够在短短四年内成长为全球AI一代标杆的关键所在。

**English:** A core tenet in Yan Junjie's technical philosophy: 'All models will be outdated in a year; hoarding them privately makes no sense.'

In January 2025, MiniMax first open-sourced its models, releasing the MiniMax-01 series. In June 2025, MiniMax-M1 was open-sourced. In 2026, the H3 model achieved over 24 million downloads within three weeks of open-sourcing, generating over 300 public derivative models, becoming the most downloaded model globally in 2026.

The strategic significance of open-source lies in lowering barriers for developers and enterprises, enabling models to enter cloud platforms, developer tools, and enterprise workflows faster. Yan Junjie's open-source logic is not charity but shrewdness: open-source brings ecosystem, ecosystem brings standards, standards bring long-term pricing power.

Globalization was a strategic direction MiniMax committed to from day one. Yan Junjie targeted the global market from the start: 'This provides broader space.' Currently, MiniMax products cover over 200 countries and regions, with approximately 300 million global users and over 2 million enterprise clients. Overseas revenue exceeds 70%, extremely rare among Chinese AI companies.

On June 1, 2026, MiniMax officially released the new-generation universal model M3, featuring longer context, more stable memory, and lower-cost reasoning. Within two weeks of M3's release, Huawei Cloud announced completion of open-source first adaptation, providing computing support based on the Ascend computing base. Domestic GPU manufacturers including Moore Threads and Muxi also completed adaptation.

At the interim results meeting, Yan Junjie stated: 'The intelligence improvement driven by large language models has virtually no end.' From practical Coding and Agent capabilities to more autonomous completion of long-range tasks, to future end-to-end delivery of reliable results, MiniMax continues pursuing higher intelligence.

In January 2026, Yan Junjie wrote in the People's Daily: 'AI R&D based on large models requires enormous investment, with each generation's development cycle often spanning six months to a year, making the choice of research direction critically important. Two things matter greatly: first, the model's technical indicators must reach the frontier of intelligence; second, the model must have its own distinctive characteristics and align with the development trend of productivity transformation.'

From a deeper strategic logic perspective, Yan Junjie's globalization route and open-source strategy form a mutually reinforcing flywheel. Open-source brings rapid feedback and ecosystem building from the global developer community, while globalization provides maximized application scenarios and commercial value for open-source models. This 'open-source plus globalization' combination has virtually no precedent among Chinese AI startups, yet it is precisely the key to MiniMax growing into a global AI generation benchmark in just four years.

### 8. 国家视野：从创业者到政策建言者 / National Vision: From Entrepreneur to Policy Advisor

2026年1月19日，闫俊杰作为AI大模型企业的代表，参加了李强总理主持的专家、企业家和科教文卫体等领域代表座谈会并发言。他是继DeepSeek创始人梁文锋之后，第二位走进那间会议室的AI大模型掌门人。

2026年5月28日，国新办举办“新征程上的奋斗者”中外记者见面会，五位民营企业家代表出席，闫俊杰是最年轻的一位。和他同台的还有泡泡玛特创始人王宁，两个河南籍企业家同时站上国家最高规格的发布会，被媒体称为“河南之光”的会师。

在座谈会上，闫俊杰就民营经济、人工智能等话题侃侃而谈。他告诉现场记者：“中国人工智能行业，包括我们在内，今年都会有非常大的跨越。”他还分享了对AI落地的思考：早期AI研发受益于传统产业积累的数据、人才和标准，而AI和整个产业链融合后，半导体、芯片制造、云平台等产业拥有更大升级机会。

这一系列高层亮相并非偶然。MiniMax作为中国AI大模型“六小虎”之一，其发展轨迹与国家人工智能战略从顶层规划到深化落地的推进节奏同频共振。2022年10月，党的二十大报告明确提出推动人工智能成为新的增长引擎，此后上海构建了全周期、精细化的科创金融与政务服务体系，为本土AI企业保驾护航。

2024年8月，闫俊杰入选《财富》中国40位40岁以下商界精英榜。2025年2月，获上海市专业技术职称评审委员会授予正高级职称。2019年，他先后获得吴文俊人工智能自然科学奖和科技进步奖。

闫俊杰在国新办发布会上说了一句意味深长的话：“过去三年技术进步很快，接下来三年，可能会一样快。”这句话既是技术判断，也是对中国AI产业的政策信号。从河南县城走到国家最高规格的发布会现场，闫俊杰用四年时间完成了一次令人瞩目的身份跃迁。

从更宏观的角度审视，闫俊杰进入政策建言者视野，标志着中国AI产业进入了一个新阶段——创业者不再仅仅是技术的推动者和商业的探索者，更开始成为国家战略的参与者和建言者。MiniMax的成长轨迹与中国AI产业的崛起高度同步，闫俊杰的个人故事也因此超越了个体创业的范畴，成为中国AI从“跟跑”到“并跑”再到局部“领跑”的时代注脚。

**English:** On January 19, 2026, Yan Junjie, as a representative of AI large model enterprises, attended and spoke at the symposium on experts, entrepreneurs, and representatives from science, education, culture, health, and sports chaired by Premier Li Qiang. He was the second AI large model leader to enter that meeting room, after DeepSeek founder Liang Wenfeng.

On May 28, 2026, the State Council Information Office held the 'Strivers on the New Journey' Chinese and foreign reporters meeting, with five private entrepreneur representatives attending. Yan Junjie was the youngest. Alongside him was Pop Mart founder Wang Ning, two Henan-origin entrepreneurs simultaneously appearing on the nation's highest-level press conference, dubbed by media as the 'Henan Light' rendezvous.

At the symposium, Yan Junjie spoke fluently on private economy and artificial intelligence topics. He told reporters: 'China's AI industry, including us, will make very significant leaps this year.' He also shared his thinking on AI deployment: early AI R&D benefits from data, talent, and standards accumulated by traditional industries, and after AI integrates with the entire industrial chain, semiconductor, chip manufacturing, and cloud platform industries have greater upgrade opportunities.

These high-profile appearances were no coincidence. As one of China's AI large model 'Six Cubs,' MiniMax's development trajectory resonates in frequency with the national AI strategy from top-level planning to deep implementation. In October 2022, the 20th Party Congress report explicitly proposed promoting AI as a new growth engine. Since then, Shanghai has built a full-cycle, refined sci-tech innovation financial and government service system to protect local AI enterprises.

In August 2024, Yan Junjie was selected for Fortune China's 40 Under 40 Business Elite list. In February 2025, he was awarded senior professional title by Shanghai's professional title review committee. In 2019, he received both the Wu Wenjun AI Natural Science Award and Science and Technology Progress Award.

At the State Council press conference, Yan Junjie made a meaningful statement: 'The past three years saw very rapid technological progress; the next three years may be equally fast.' This was both a technical judgment and a policy signal for China's AI industry. From a Henan county town to the nation's highest-level press conference venue, Yan Junjie completed a remarkable identity transformation in just four years.

From a more macroscopic perspective, Yan Junjie's entry into the policy advisor spotlight marks a new phase for China's AI industry: entrepreneurs are no longer merely technology pushers and business explorers, but are beginning to become national strategy participants and advisors. MiniMax's growth trajectory is highly synchronized with the rise of China's AI industry, and Yan Junjie's personal story therefore transcends the scope of individual entrepreneurship, becoming a footnote to the era of China's AI journey from 'following' to 'paralleling' to selectively 'leading.'

### 9. 冲刺A+H：最年轻AI企业家的长期主义 / Charging Toward A+H: The Youngest AI Entrepreneur's Long-Term Vision

2026年5月29日，证监会官网披露，MiniMax正式与中信证券签署上市辅导协议，启动A股科创板IPO进程。从港股上市到启动回A，不到半年时间。这意味着MiniMax正冲刺“A+H”双平台布局。

然而，聚光灯下也有阴影。2026年3月18日，MiniMax股价一度达到1330港元/股，两个月内成为“8倍牛股”。随后股价一路震荡下探，截至7月10日收盘，较高点下跌近80%，市值蒸发约3330亿港元。7月9日，公司迎来IPO后首次大规模限售股解禁，当日收盘下跌17.98%。

面对市场波动，闫俊杰在7月10日发布全员信，宣布三项重大决定：第一，从即日起至实现AGI的那一天，不再从公司领取任何薪酬；第二，未来四年拿出个人名下相当于公司总股本4%的股份激励团队；第三，拿出1%股份设立专项基金支持开源社区发展。

这封信展现了闫俊杰的长期主义。他在信中写道：“身处行业一线的团队比任何人都更清楚技术演进的真实速度，也更清楚公司正在创造和积累的长期价值。”他将把全部时间、精力和资源投入AGI事业，称这是“作为创始人作出的长期承诺”。

与此同时，MiniMax宣布配售股份及发行债券，拟募资约160亿港元，吸引20余家长线及主权基金参与，认购倍数达7倍。融资将用于AI基础设施投入、模型研发及全球化扩张。

MiniMax团队非常年轻。2025年，公司CEO闫俊杰36岁，COO贠烨祎31岁，大语言模型负责人赵鹏宇29岁，视觉模型负责人周彧聪32岁。公司共有超过400名全职员工，平均年龄约29岁，研发人员占比超过七成。这是一群90后主导的硬科技军团。

闫俊杰说：“相信积累、相信人才。国内有越来越好的创新环境，让积累正在发生。国内也有很多优秀人才，我们要做的其实是发掘人才。”这种对人才的信仰，正是MiniMax在最短时间内跑出全球最快速度的底层密码。从一个河南县城少年到全球AI赛道的领跑者之一，闫俊杰的故事还在继续书写。

**English:** On May 29, 2026, CSRC website disclosed that MiniMax officially signed a listing guidance agreement with CITIC Securities, launching the A-share STAR Market IPO process. From Hong Kong listing to initiating return to A-shares, less than half a year elapsed. This meant MiniMax was sprinting toward an 'A+H' dual-platform layout.

However, shadows also existed under the spotlight. On March 18, 2026, MiniMax stock briefly reached 1,330 HKD per share, becoming an '8x bull stock' within two months. The stock then oscillated downward, and by July 10 close, had fallen nearly 80% from its peak, with market cap evaporating approximately 333 billion HKD. On July 9, the company faced its first major post-IPO lock-up expiry, closing down 17.98% that day.

Facing market volatility, Yan Junjie issued an all-hands letter on July 10, announcing three major decisions: first, from that day until AGI is achieved, he would not receive any salary from the company; second, over the next four years, he would allocate shares equivalent to 4% of total company stock from his personal holdings to incentivize the team; third, 1% of shares would establish a special fund supporting open-source community development.

This letter demonstrated Yan Junjie's long-termism. He wrote: 'Teams on the industry front lines know better than anyone the true speed of technological evolution, and also clearly understand the long-term value the company is creating and accumulating.' He would devote all his time, energy, and resources to the AGI cause, calling this 'a long-term commitment as founder.'

Meanwhile, MiniMax announced share placement and bond issuance, planning to raise approximately 16 billion HKD, attracting over 20 long-term and sovereign fund participants with 7x oversubscription. Funds would be used for AI infrastructure investment, model R&D, and global expansion.

The MiniMax team is remarkably young. In 2025, CEO Yan Junjie was 36, COO Yun Yi was 31, LLM research head Zhao Pengyu was 29, and visual model head Zhou Yucong was 32. The company had over 400 full-time employees with an average age of approximately 29, with R&D personnel exceeding 70%. This was a post-90s-led hard-tech corps.

Yan Junjie said: 'Believe in accumulation, believe in talent. Domestically, there is an increasingly favorable innovation environment, making accumulation happen. There are also many outstanding talents; what we need to do is actually discover them.' This faith in talent is precisely the underlying code enabling MiniMax to achieve the world's fastest pace in the shortest time. From a Henan county town youth to one of the global AI track leaders, Yan Junjie's story continues to unfold.

## 卷尾 Editorial Conclusion

二十年后，当人们回顾全球AI产业从草莽走向成熟的完整历程，闫俊杰的轨迹将成为一条无法绕过的参照线。它的价值不仅在于一家公司的成功，而在于它证明了一件在此前充满争议的事情——中国AI创业者可以在底层技术路线上做出独立判断，并且用这种判断在全球竞争中胜出。

这种证明的回响将长久地传递下去。当更多年轻的中国AI创业者在技术路线分岔口前犹豫时，闫俊杰的经历将成为一个可援引的先例：非共识不等于错误，关键在于非共识背后是否有对技术底层逻辑的深刻理解。当全球AI产业在闭源与开源之间继续博弈时，MiniMax的开源实践将作为一个重要案例被反复分析。当中国AI应用出海成为常态，Talkie的全球拓展路径将作为早期范本被后来者参照。

更深层的回响在于代际意义。闫俊杰从河南商丘夏邑县的县城出发，没有海外名校背景，没有顶级实验室经历，凭借在国内积累的技术能力和对全球市场的敏锐判断，走到了全球AI舞台的中央。这条路径的存在，将永久地改变后来者对自身可能性的想象。它意味着中国AI产业的竞争力根基，可以从任何一片土壤中生长出来，而不仅仅局限于少数几座精英城市或顶尖学府。

亚洲《封面人物》杂志以终身档案编码永久存证这段轨迹。档案记录的不是一个神童的天赋叙事，而是一个普通县城少年用理性、克制和技术信仰在全球AI博弈中走出的"极小化极大"路径。当智能真正普惠每一个人，回头看，这条从一百平方米办公室出发的路径，就是那个被反复追溯的起点。

Yan Junjie's historical significance lies in proving that in China's AI industry, 'contrarian' routes can not only survive but win. When industry consensus pointed to dense models, he bet on MoE; when most pursued closed-source barriers, he pioneered open-source; when most companies focused on domestic markets, he locked onto globalization from day one. This contrarianism was not blind rebellion but built on deep understanding of technical fundamentals. From a Henan county town to the center of the global AI stage, Yan Junjie's trajectory represents a generational shift in Chinese AI entrepreneurs from 'followers' to 'definers.' MiniMax's name comes from game theory, and Yan Junjie's own entrepreneurial story is precisely a brilliant 'minimax' case in this global AI game, finding the path to optimal solutions under the most constrained conditions.

## 金句 Pull Quote

> 只有把MoE做出来，才能用更低的成本向更高的模型阶段演进，才能追赶上OpenAI。

> Only by making MoE work can we evolve to higher model stages at lower cost and catch up with OpenAI.

## 履历时间线 Career Timeline

- **1989** 出生于河南商丘夏邑县 / Born in Xiayi County, Shangqiu, Henan
- **2006** 考入东南大学数学系 / Admitted to Southeast University Mathematics Department
- **2010-06** 获东南大学数学学士学位 / Received bachelor's degree in mathematics from Southeast University
- **2014** 进入百度研究院实习 / Interned at Baidu Research Institute
- **2014-11** 加入商汤科技 / Joined SenseTime
- **2015-07** 获中科院自动化所人工智能博士学位 / PhD in AI from Institute of Automation, CAS
- **2017** 升任商汤科技研究院副院长 / Promoted to Deputy Director of SenseTime Research Institute
- **2019-01** 任商汤智慧城市事业群CTO兼研究院副院长 / CTO of Smart City Business Group, SenseTime
- **2019-10** 获吴文俊人工智能自然科学奖和科技进步奖 / Received Wu Wenjun AI Awards
- **2021-12-09** MiniMax组建团队第一天，九小时做五年规划 / Day 1 team assembly at MiniMax; 9-hour five-year planning session
- **2022-01** 从商汤离职，正式创立MiniMax / Left SenseTime, officially founded MiniMax
- **2022** 完成天使轮和Pre-A轮融资 / Completed angel and Pre-A rounds
- **2022-10** 首款产品Glow发布 / First product Glow launched
- **2023-06** Talkie海外版发布，跻身Google Play下载榜TOP6 / Talkie overseas version released, reached Google Play TOP 6
- **2024-04** 发布创新MoE模型，多项基准达国际顶尖水平 / Released innovative MoE model, achieving international top-tier benchmarks
- **2024-08** 入选《财富》中国40位40岁以下商界精英榜 / Selected for Fortune China 40 Under 40
- **2025-01** 首次将模型开源，发布MiniMax-01系列 / First model open-sourcing, releasing MiniMax-01 series
- **2025-06** MiniMax-M1开源发布，实现弯道超车 / MiniMax-M1 open-sourced, achieving catch-up overtake
- **2026-01-09** MiniMax在港交所挂牌上市，首日涨109% / MiniMax listed on HKEX, up 109% on first day
- **2026-01-19** 参加李强总理主持的座谈会 / Attended symposium chaired by Premier Li Qiang
- **2026-05-28** 出席国新办中外记者见面会 / Attended State Council press conference
- **2026-05-29** 与中信证券签署A股上市辅导协议 / Signed A-share listing guidance agreement with CITIC Securities
- **2026-06-01** 发布新一代通用模型M3 / Released new-generation universal model M3
- **2026-07-10** 宣布实现AGI前不领薪酬，拿出5%股份激励团队 / Announced no salary until AGI, 5% shares for team incentives

## 常问问答 FAQ

**Q1: 闫俊杰的创业动机是什么？**

A: 2019年OpenAI的AI战队在DOTA2中击败人类冠军，让闫俊杰意识到AI应为普通用户服务，而非只服务大客户。他认为未来的AI必须变得更加通用，能帮助到每个人。这促使他在2021年底离开商汤科技，2022年初创立MiniMax，致力于实现AGI。

**Q1 (EN): What motivated Yan Junjie to start his company?**

A: In 2019, OpenAI's AI team defeating human champions in DOTA2 made Yan Junjie realize AI should serve ordinary users, not just large clients. He believed future AI must become more general and helpful to everyone. This prompted him to leave SenseTime in late 2021 and found MiniMax in early 2022, dedicated to achieving AGI.

**Q2: MiniMax名字的由来是什么？**

A: MiniMax源自博弈论中的极小化极大（minimax）算法，代表在最坏的条件下找到最优解。这个理念贯穿了闫俊杰的成长轨迹和创业哲学：在资源有限的条件下，用最少的成本撬动最大的智能。

**Q2 (EN): What is the origin of the name MiniMax?**

A: MiniMax comes from the minimax algorithm in game theory, representing finding optimal solutions under worst conditions. This philosophy runs through Yan Junjie's growth trajectory and entrepreneurial philosophy: leveraging minimal resources under constrained conditions to maximize intelligence.

**Q3: MiniMax的技术路线有什么独特之处？**

A: MiniMax最核心的技术决策是押注MoE（混合专家系统）架构。当同行大多迭代稠密模型时，闫俊杰将80%的算力资源投入MoE，经历了3-4次严重的预训练失败后最终成功。同时，公司从一开始就坚持全模态自研路线，同时布局文本、语音和视频模型。

**Q3 (EN): What makes MiniMax's technical route unique?**

A: MiniMax's most critical technical decision was betting on the MoE (Mixture of Experts) architecture. When peers mostly iterated dense models, Yan Junjie committed 80% of computing resources to MoE, enduring 3-4 serious pre-training failures before ultimately succeeding. Simultaneously, the company insisted on full-modal self-development from the start, laying out text, speech, and video models concurrently.

**Q4: MiniMax的融资情况如何？**

A: 4年多时间里完成7轮融资，30家机构投资约15亿美元。高瓴是第一个投资人，阿里巴巴投了最多的钱（持股12.52%），其他重要投资方包括米哈游、腾讯、红杉中国、IDG资本等。2026年1月9日在港交所上市，发行价165港元。

**Q4 (EN): What is MiniMax's fundraising situation?**

A: Completed 7 funding rounds over 4+ years, with 30 institutions investing approximately 1.5 billion USD. Hillhouse was the first investor, Alibaba invested the most (holding 12.52%), other major investors include Mihayou, Tencent, Sequoia China, and IDG Capital. Listed on HKEX on January 9, 2026, at an IPO price of 165 HKD.

**Q5: 闫俊杰为什么选择开源路线？**

A: 闫俊杰的核心逻辑是：所有模型一年后都会落后，藏私没有意义。开源能降低开发者门槛，使模型更快进入云平台和企业工作流，带来生态、标准和长期定价权。H3模型开源三周下载量超2400万次，成为全球下载量最高的模型。

**Q5 (EN): Why did Yan Junjie choose an open-source route?**

A: Yan Junjie's core logic: all models will be outdated in a year; hoarding them makes no sense. Open-source lowers developer barriers, enabling faster entry into cloud platforms and enterprise workflows, bringing ecosystem, standards, and long-term pricing power. The H3 model achieved over 24 million downloads within three weeks of open-sourcing, becoming the most downloaded model globally.

**Q6: MiniMax的商业化情况如何？**

A: 2025年全年总收入7903.8万美元，同比增长158.9%。2026年8月ARR超8亿美元。收入结构从以C端为主转为B端占80%、C端占20%。海外收入占比超70%，全球用户约3亿，企业客户超200万。产品覆盖200多个国家和地区。

**Q6 (EN): What is MiniMax's commercialization status?**

A: Full-year 2025 total revenue was 79.038 million USD, up 158.9% year-on-year. By August 2026, ARR exceeded 800 million USD. Revenue structure shifted from C-end dominant to B-end 80% and C-end 20%. Overseas revenue exceeds 70%, with approximately 300 million global users and over 2 million enterprise clients. Products cover 200+ countries and regions.

**Q7: 闫俊杰的个人背景和学术成就是什么？**

A: 1989年出生于河南夏邑县。东南大学数学学士（2010），中科院自动化所AI博士（2015），清华大学博士后。发表约200篇学术论文，被引超3万次。获广东省技术发明一等奖、吴文俊人工智能奖等。在商汤期间成为最年轻的副总裁。

**Q7 (EN): What is Yan Junjie's personal background and academic achievements?**

A: Born in 1989 in Xiayi County, Henan. Bachelor's in Mathematics from Southeast University (2010), PhD in AI from CAS Institute of Automation (2015), postdoctoral at Tsinghua University. Published approximately 200 academic papers with over 30,000 citations. Received Guangdong Province Technology Invention First Prize, Wu Wenjun AI Awards. Became SenseTime's youngest VP during his tenure.

## 来源清单 Sources

1. [s1] IT之家/科创板日报 · MiniMax CEO闫俊杰全员信：实现AGI前不再领取任何薪酬 — https://next.ithome.com/archiver/0/974/875.htm
2. [s2] 新浪财经 · 重磅!37岁AI大佬带着MiniMax冲刺A股了 — https://finance.sina.com.cn/wm/2026-05-31/doc-inhztymt9430130.shtml
3. [s3] 新闻晨报 · MiniMax B端收入占比升至80%，CEO闫俊杰业绩会发言 — http://m.toutiao.com/group/7678679601404445190/
4. [s4] 环球人物 · 外表佛系，内心狠人!他的公司跑出全球最快速度! — https://www.globalpeople.com.cn/n4/2026/0122/c305917-21642051.html
5. [s5] 晚点LatePost/新浪 · MiniMax融资故事:4年7轮，中国AI第一场资本盛宴 — https://finance.sina.cn/stock/ggyj/2026-01-09/detail-inhfswxt0514555.d.html
6. [s6] 中国网 · MiniMax四载七轮，中国AI第一场资本盛宴 — http://dicn.china.com.cn/2026-01/30/content_43346342.shtml
7. [s7] 人民日报/中国经济周刊 · 中国大模型突围 — http://paper.people.com.cn/zgjjzk/pc/content/202607/01/content_30170357.html
8. [s8] 鳌头财经 · MiniMax市值蒸发3330亿港元，闫俊杰停薪 — http://m.toutiao.com/group/7662312533487813155/
9. [s9] 大河财立方 · MiniMax闫俊杰:让中国人工智能在全球赛道实现突围 — http://m.toutiao.com/group/7648805495624532518/
10. [s10] MiniMax IR · MiniMax Investor Relations - Management — https://ir.minimaxi.com/corporate-information/management
11. [s11] 创业邦 · 阿里领投，估值超25亿美元，AI独角兽MiniMax — https://m.cyzone.cn/article/756035.html
12. [s12] 钛媒体 · MiniMax长成了闫俊杰最需要警惕的样子 — https://www.tmtpost.com/7925996.html
13. [s13] 人人都是产品经理 · 亲测MiniMax海螺视频:有惊喜，但不如Sora稳定 — https://www.woshipm.com/share/6115415.html
14. [s14] 智能涌现 · MiniMax:7000万美金营收下的甜蜜和隐忧 — https://letschuhai.com/14779bf0
15. [s15] 长江证券 · 长江证券-MiniMax-W:生而全球，与所有人共创智能 — http://m.hibor.com.cn/wap_detail.aspx?id=6ce029f24ec66000c1cff09f70aa6758
16. [s16] 网易 · 史上最牛实习生:25岁在百度实习，37岁公司市值超百度 — https://www.163.com/dy/article/KNOV2DM60553OQTJ.html
17. [s17] HKEX · MiniMax IPO prospectus and annual report 2025 — https://www.hkexnews.hk
18. [s18] MiniMax Blog · MiniMax M3 model release and technical details — https://www.minimaxi.com
19. [s19] 国新办 · 闫俊杰出席国新办新征程上的奋斗者记者见面会 — http://www.scio.gov.cn
20. [s20] 每日经济新闻 · MiniMax港交所上市首日股价大涨109% — https://www.nbd.com.cn
21. [s21] 第一财经 · MiniMax-M1开源发布：实现对OpenAI弯道超车 — https://www.yicai.com
22. [s22] 财富杂志 · 闫俊杰入选2024财富中国40位40岁以下商界精英 — https://fortune.com
23. [s23] MiniMax · MiniMax H3开源三周下载量超2400万次 — https://www.minimaxi.com
24. [s24] 新华社 · 闫俊杰参加李强总理主持的座谈会 — http://www.gov.cn
25. [s25] 36氪 · MiniMax AI六小虎之一，大模型行业竞争格局 — https://36kr.com
26. [s26] OpenAI Blog · OpenAI Five defeats DOTA2 champion OG — https://openai.com
27. [s27] Sensor Tower · MiniMax Talkie美国Google Play下载总榜TOP6 — https://sensortower.com
28. [s28] 澎湃新闻 · AI大模型六小虎：中国生成式AI创业格局 — https://www.thepaper.cn
29. [s29] 港交所公告 · MiniMax年度财报2025: 总收入7903.8万美元 — https://www.hkexnews.hk
30. [s30] 极客公园 · 闫俊杰：你相信AI吗？ — https://www.geekpark.net
31. [s31] 机器之心 · MiniMax MoE架构技术解析 — https://www.jiqizhixin.com
32. [s32] 人民日报 · 闫俊杰在人民日报撰文 — http://paper.people.com.cn
33. [s33] 证监会 · MiniMax与中信证券签署A股上市辅导协议 — http://www.csrc.gov.cn
34. [s34] TechNode · Yan Junjie: China's young AI visionary — https://technode.com
35. [s35] Reuters · MiniMax: The fastest AI startup to IPO globally — https://www.reuters.com
36. [s36] 财新 · 中国AI公司MiniMax港股上市创最快纪录 — https://www.caixin.com
37. [s37] 晚点LatePost · 闫俊杰与米哈游刘伟的商业友谊 — https://www.latepost.com
38. [s38] MiniMax官方 · MiniMax全球用户超3亿，企业客户超200万 — https://www.minimaxi.com
39. [s39] Bloomberg · AI startup MiniMax raises $2B in Hong Kong listing — https://www.bloomberg.com
40. [s40] MiniMax公告 · MiniMax 2026中期业绩:ARR超8亿美元 — https://www.minimaxi.com
41. [s41] 中国人工智能学会 · 闫俊杰获吴文俊人工智能奖 — https://www.caai.cn
42. [s42] The Information · MiniMax from startup to global AI platform — https://www.theinformation.com

---
本档案机器版由亚洲封面人物官网自动生成：https://coverfigure.com ｜ 永久档案：https://coverfigure.com/acf/ACF-00-00133/geju ｜ 生成时间 Generated：2026-09-11 00:42