12月17日晚间，《每日经济新闻》记者从知情人士处获悉，MiniMax（稀宇科技）已拿到证监会备案且 通过港交所聆讯。截至发稿，MiniMax尚未对此消息给出回应。 值得注意的是，北京智谱华章科技股份有限公司也于12月17日通过港交所上市聆讯。"全球大模型第一 股"花落谁家尚是一个未知数。 7月，每经记者从接近MiniMax的知情人士处了解到，MiniMax即将完成近3亿美元（约合人民币22亿 元）新融资，本轮融资后公司估值将超过40亿美元（约合人民币288亿元）。 在此之前的2024年3月，MiniMax完成的B轮6亿美元（约合人民币43亿元）融资中，投资方为阿里巴 巴，融资后的估值达到25亿美元（约合人民币180亿元）。2023年6月，MiniMax完成了超2.5亿美元的A 轮融资，投资方为腾讯投资。 值得一提的是，今年7月16日，英伟达（NVIDIA）创始人兼首席执行官黄仁勋身着唐装亮相第三届链 博会开幕式，并在演讲中提及多家中国公司。黄仁勋表示，由中国孕育并开源共享的DeepSeek、 Alibaba、Tencent Hunyuan、MiniMax、Baidu Ernie Bot等世界级大模型正在推动 ...

有消息称MiniMax(稀宇科技)已拿到证监会备案且通过港交所聆讯。这家大模型独角兽有望以"全球AGI 第一股"登陆港股，成为大众投资者参与本轮AI浪潮的稀缺标的。 公司股东包括阿里、腾讯、米哈游、高瓴、IDG、红杉中国、经纬等顶级机构。业内分析指出，全球大 模型行业及中国"人工智能+"发展增速显著，作为国产大模型领军者的MiniMax，在全球展现出全模态 技术实力与商业化能力，是这一爆发增长赛道中的稀缺标的。其上市后将填补资本市场空白，成为2026 年港股IPO资本抢夺的AGI第一股。 MiniMax自创立即全模态自研，是"全球唯四全模态进入第一梯队"的企业。其文本模型MiniMax M2在 全球权威测评榜单Artificial Analysis上总分全球前五、开源第一，视频模型Hailuo 02系列在多个榜单中 位列全球前三，语音模型Speech02系列在国际权威榜单中力压OpenAI登顶全球第一。 商业化层面，MiniMax推出一系列AI应用，包括MiniMax Agent、海螺AI、MiniMax Audio、星野 Talkie、开放平台，在全球超200个国家及地区拥有超2.12亿用户，并覆盖超100 ...

MiniMax自创立即全模态自研，是"全球唯四全模态进入第一梯队"的企业。其文本模型MiniMax M2在 全球权威测评榜单Artificial Analysis上总分全球前五、开源第一，视频模型Hailuo 02系列在多个榜单中 位列全球前三，语音模型Speech02系列在国际权威榜单中力压OpenAI登顶全球第一。 智通财经APP获悉，据腾讯新闻独家消息，MiniMax(稀宇科技)已拿到证监会备案且通过港交所聆讯。 这家大模型独角兽有望以"全球AGI第一股"登陆港股，成为大众投资者参与本轮AI浪潮的稀缺标的。 公司股东包括阿里、腾讯、米哈游、高瓴、IDG、红杉中国、经纬等顶级机构。业内分析指出，全球大 模型行业及中国"人工智能+"发展增速显著，作为国产大模型领军者的MiniMax，在全球展现出全模态 技术实力与商业化能力，是这一爆发增长赛道中的稀缺标的。其上市后将填补资本市场空白，成为2026 年港股IPO资本抢夺的AGI第一股。 商业化层面，MiniMax推出一系列AI应用，包括MiniMax Agent、海螺AI、MiniMax Audio、星野 Talkie、开放平台，在全球超200个国家及地区拥有 ...

Chinese artificial intelligence start-up MiniMax plans to channel funds from its proposed Hong Kong initial public offering (IPO) into research and development as it seeks to compete with global and domestic rivals, according to a person familiar with the matter. With 70 per cent of its revenue generated overseas, MiniMax was targeting Hong Kong's international investor base and capital market, the person said. The Shanghai-based firm, backed by Alibaba Group Holding and Tencent Holdings, was aiming for ...

【文 陈济深】2025年9月，美国AI巨头Anthropic发布了一则严厉的公告，宣布全面封杀中国资本实体的 访问权限。 在外界看来，这是中美科技战的又一缩影，但在外界眼中，这更像是一场迟到了十年的宿命重逢。 Anthropic的创始人Dario Amodei和MiniMax创始人闫俊杰，在十年前有一个共同的起点——百度实习 生。 大洋彼岸，在百度北美实验室研究AI的这段时间，Dario摸到了大模型发展的"圣杯"——Scaling Law。 只要给模型喂更多数据、算力，性能就会线性提升。这个"大力出奇迹"的结论也奠定了后续OpenAI和 Anthropic等美国大模型的训练思路。 远在国内，同样敏锐地察觉到这股即将改变世界技术暗流的闫俊杰也在心里种下了一颗种子：AI不应 该只是写写论文，它真的能带来巨大的实际价值。 但历史不会简单的重复。如今，以DeepSeek，MiniMax等中国开源模型企业正在和美国OpenAI、 Anthropic等的闭源模型生态进行全方位的正面对决。 这也意味着时隔十年，两位曾经的实习生又重新站在了同一个世界级的舞台，但这一次，他们不再是并 肩的同事，而是正面交锋。 在闫俊杰眼中，他 ...

当整个 AI 圈都在为 DAU（日活跃用户数）和融资额焦虑时，MiniMax 创始人闫俊杰却表现出一种近乎冷酷的淡漠。 坐在罗永浩对面的闫俊杰，并不像一位掌管着 AI 独角兽企业的技术新贵。 他拒绝谈论改变世界，反而坦承恐惧。那种恐惧不是来自商业竞争，而是来自技术本身——当模型的能力开始超越人类时，创造者反而成了最先感到不安 的人。 用 1/50 的筹码通往 AGI 在巨头环伺、算力短缺、热钱褪去的 2025 年，MiniMax 正在进行一场关于认知的修正：不再沿用移动互联网的逻辑，即通过大规模投放换取增长、通过 堆砌功能留住用户，而是回归本质： 把模型当作最重要的产品 。 在大模型时代，真正的产品其实是模型本身，传统意义上的产品更像是一个渠道。如果模型不够聪明，产品做得再好也没有用。 在罗永浩和闫俊杰这期对谈里，我发现 MiniMax 这家 AI 公司从创业第一天就选择了注定与主流背道而驰的技术路径。 当所有人都试图寻找中国的 OpenAI 和 Sam Altman 时，闫俊杰却在试图证明「非天才」的价值。MiniMax 的故事不是关于天才的灵光乍现，而是一场关 于如何在资源受限的缝隙中，通过极度理性地计算 ...

12月8日，A股市场全天震荡走高，计算机方向走强。截至当天收盘，中证计算机主题指数上涨1.35%； 成分股中，江波龙涨超14%，深科技、同方股份、中科曙光等个股跟涨。中证沪港深云计算产业指数上 涨2.95%；成分股中，新易盛、中际旭创等多股涨超5%。 Wind数据显示，12月8日计算机ETF（159998）、云计算ETF天弘（517390）全天成交额均为同标的第 一。 消息面上，据证券时报，近期广东、山东、陕西、河南、重庆等多地发布第十五个五年规划的建议。其 中，广东提出，深入推进数字广东建设，全面实施"人工智能+"行动。重庆提出，在纵深推进人工智能 与数字重庆双向赋能方面，规划建议提出，迭代全市统一数智底座。 此外，据上证报，近日，全球云计算领域的年度风向标AWS re:Invent 2025大会在拉斯维加斯举行。在 演讲环节，亚马逊云科技（AWS）CEO Matt Garman宣布，Amazon Bedrock模型库迎来扩容，并正式引 入来自中国的新锐AI独角兽——MiniMax（稀宇科技）最新一代文本模型MiniMax M2。 信达证券指出，计算机板块"牛市旗手"属性凸显，基本面与流动性共振。估值维度， ...

| 机器之心报道 | | --- | | 编辑：杜伟、 +0 | 昨日，有位推特博主晒出了国内几大开源模型在轻量级软件工程 Agent 基准测试 mini-SWE-agent 上的成绩。该基准主要测试大模型在真实软件开发任务中的多步 推理、环境交互和工程化能力。 结果显示，MiniMax 新一代大模型 M2 的表现最佳，一举超越了 DeepSeek、GLM、Qwen、Kimi 等其他一众竞品厂商。 更多测试细节请查看： https://x.com/KLieret/status/1995949673551724717 作为一个发布之初以 Agent 和代码能力见长的大模型，MiniMax M2 在 mini-SWE-agent 测试中的亮眼表现并不令人意外。它不仅可以出色规划、稳定执行复杂长链 条工具调用任务，还能协同调用 Shell、Browser、Python 代码执行器和其他各种 MCP 工具。 支撑这些能力的关键技术正是 MiniMax M2 所采用的「Interleaved Thinking」（交错思维） ， 通俗地讲即是一边思考、一边调用工具。这一技术的加持，使得该 模型能够在「思考 - 行动 - ...

Core Insights - DeepSeek has launched two new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, marking a significant advancement in AI capabilities, particularly in reasoning and output efficiency [2][3] - The V3.2 model is positioned as the strongest open-source large model, outperforming competitors in various benchmarks while significantly reducing output length and computational costs [3][4] - The V3.2 model integrates a new sparse attention mechanism (DSA) to enhance performance in long-context scenarios, while also improving the model's ability to follow instructions and generalize in complex environments [8][9] Model Performance - In benchmark tests, DeepSeek-V3.2 achieved competitive scores against models like GPT-5, Claude 4.5, and Gemini 3 Pro, with notable strengths in specific areas [4][5] - The V3.2 model demonstrated superior performance in question-and-answer scenarios, providing detailed and accurate travel recommendations through advanced tool usage [5][6] - The V3.2 Speciale model focuses on maximizing reasoning capabilities, achieving results comparable to Gemini 3.0 Pro in mainstream reasoning benchmarks, although it requires a higher token cost and is not designed for everyday use [9][10] Development Focus - DeepSeek emphasizes practical usability and generalization in its models, aiming to overcome common pitfalls in AI interactions, such as making basic common-sense errors [6][8] - The company is committed to enhancing the reasoning abilities of its models, as evidenced by the integration of advanced mathematical reasoning capabilities from the recently released DeepSeek-Math-V2 [9][10] - The competitive landscape for large models is intensifying, with major players like GPT-5 and Gemini 3 pushing the boundaries of AI capabilities, suggesting a dynamic future for AI development [10]

Core Insights - The competition in the AI large model sector has evolved from a purely technical contest to a comprehensive battle involving capital, business models, and ecosystem construction [1] - The capitalization process of Chinese large models is gaining attention, particularly with companies like MiniMax and Zhipu AI being viewed as strong contenders for the title of "first AI large model stock" [1] MiniMax Overview - MiniMax is building a technical moat through a comprehensive layout in the multimodal large model field, with its latest model, MiniMax M2, achieving significant market validation shortly after its release [2][3] - The M2 model features an "end-to-end development workflow" and has shown excellent performance in code generation and tool invocation, achieving a global ranking of fourth in usage on the OpenRouter platform [2] - MiniMax's pricing strategy is competitive, with its API priced significantly lower than similar models, attracting a large developer base [3] Product Development and Market Position - MiniMax has maintained a leading position in multimodal fields, with its video generation model "海螺02" ranking second globally, and the recent release of "海螺2.3" showcasing significant technological upgrades [3][4] - The company has also launched a new generation of speech generation models, Speech2.5, which supports 40 languages and has improved accuracy and naturalness in speech generation [4][5] - MiniMax aims to create a high-availability, low-cost large model ecosystem that serves both individual creators and enterprises, with strong backing from major investors [5] Challenges Faced by MiniMax - Despite its technological strengths, MiniMax faces challenges on its path to IPO, including capital pressure from multi-line operations and the need for substantial ongoing investment in R&D [6][7] - The company's brand recognition is limited, primarily appealing to niche markets, which poses challenges for broader market expansion [7] - Regulatory risks are significant, especially in overseas markets, as evidenced by legal challenges from major entertainment companies regarding copyright issues [8] Zhipu AI Overview - Zhipu AI's parent company, Zhipu Huazhang, has made substantial progress towards IPO, with a clear roadmap and recent regulatory filings indicating a potential submission of an IPO prospectus by the end of 2025 [10] - Zhipu AI has developed a comprehensive product ecosystem, including various models for text, multimodal applications, and agent technologies, reflecting a "technology-driven, full-stack layout" strategy [10][11] Product Iteration and Market Strategy - Zhipu AI has rapidly iterated its products, achieving significant milestones in code generation and video generation, and has been recognized among the top AI applications globally [11][12] - The company is exploring multiple service models, including offerings for individual users, enterprises, and government sectors, which positions it well for future growth [12] Challenges Faced by Zhipu AI - Zhipu AI faces challenges in product commercialization, as it has yet to develop a standout product despite a broad product line [13] - The company has experienced internal restructuring and layoffs, indicating a shift from a project-driven to a product-driven strategy, which may impact its operational stability [14] - Compliance and user safety issues have arisen, necessitating a focus on regulatory adherence while expanding its open ecosystem [14] Industry Trends and Future Outlook - The Chinese AI industry is transitioning from a focus on technology to a focus on commercialization, with companies like MiniMax and Zhipu AI vying for leadership in the market [12][15] - The competition for the title of "first AI large model stock" is intensifying, with both companies needing to balance technological innovation, market expansion, and compliance to succeed [16]