IT之家 2 月 26 日消息，Anthropic 当地时间 25 日宣布收购 PC 智能体应用领域初创企业 Vercept，后者 的 Vy 服务将于 30 天后关闭。 这是 Anthropic 的第二笔公开收购交易。此外IT之家注意到，Vercept 的联合创始人之一 Matt Deitke 去 年就以 2.5 亿美元的高薪合同加盟了 Meta。 Anthropic 表示，来自 Vercept 的人才将支持该公司增强 AI 的计算机使用能力，提升 Claude 助手在实 时应用程序多任务处理方面的表现。 ...

Group 1 - The company Jueyue Xingchen is considering an IPO on the Hong Kong Stock Exchange, aiming to raise approximately $500 million (around 3.4 billion RMB) [2] - Jueyue Xingchen was founded in 2023 by former Microsoft global vice president Jiang Daxin and has released the Step series of general large model matrices, covering comprehensive capabilities from language to multimodal and reasoning [2] - The company has completed two rounds of financing, with the latest round in January raising over 5 billion RMB, setting a record for the largest single financing in China's large model sector in 2025 [2] Group 2 - Under the leadership of Chairman Yin Qi, the company released its strongest open-source base model, Step 3.5 Flash, which approaches the capabilities of closed-source models in agent scenarios and mathematical tasks [3] - Jueyue Xingchen may become the third large model enterprise to be listed on the Hong Kong Stock Exchange [3]

OpenAI said Tuesday it has appointed Arvind KC as chief people officer, tapping the former Roblox Corp. executive to oversee hiring, orientation and employee development as the ChatGPT maker continues to expand. KC will report to Chief Strategy Officer Jason Kwon, according to an OpenAI spokesperson. The artificial intelligence startup is currently raising a funding round expected to bring in more than $100 billion, and has recently taken steps toward a public listing. KC’s hire follows turnover in the role ...

智通财经APP获悉，MINIMAX(00100)盘中涨超6%，截至发稿，涨5.64%，报796港元，成交额4.17亿港 元。 此外，高盛研报指出，MINIMAX作为全球顶尖的AI模型公司之一，正处于高速增长阶段。其最大的亮 点在于高度全球化的收入结构（70%的收入来自海外）以及全面的多模态产品线（涵盖文本、视频、音 频、音乐和图像）。高盛指出，MiniMax在文本/代码、多模态以及AI智能体/数字劳动力领域的广阔全 球潜在市场（TAM）中占据了极佳的竞争位置。 消息面上，2月26日早，Minimax发文介绍MiniMax Agent Expert的功能升级，同步介绍MaxClaw。据介 绍，在Expert 2.0中进一步优化了专家Agent的创建体验。用户不需要考虑Skill、SubAgent、MCP的配 置，以及提示词的结构编排，只需用自然语言描述任务目标或能力需求，Agent会根据目标完成SOP梳 理、工具编排与能力配置。与Expert 2.0升级同步，Minimax上线MaxClaw，基于OpenClaw构建的云端 AI助手，直接集成在MiniMax Agent网页端，为用户在云端部署并运行OpenCl ...

据报道，AI初创公司阶跃星辰(StepFun)正考虑在香港上市，计划筹集大约5亿美元(约39亿港元)。报道 引述知情人士称，阶跃星辰已与潜在顾问就股权出售事宜进行磋商，最早可能在今年上市，其发行规模 和上市时间等细节可能有所变化。 阶跃星辰由前微软全球副总裁姜大昕于2023年创办，主力研发基础大模型及推动"AI+终端"战略。今年 1月，据媒体报道，阶跃星辰完成逾50亿元人民币的B+轮融资，投资者包括上海国投先导基金、国寿股 权、浦东创投等产业投资者，腾讯(00700)、启明创投、五源资本等旧股东跟投。 ...

阶跃星辰由前微软全球副总裁姜大昕于2023年创办，主力研发基础大模型及推动"AI+终端"战略。今年 1月，据媒体报道，阶跃星辰完成逾50亿元人民币的B+轮融资，投资者包括上海国投先导基金、国寿股 权、浦东创投等产业投资者，腾讯(00700)、启明创投、五源资本等旧股东跟投。 智通财经APP获悉，据报道，AI初创公司阶跃星辰(StepFun)正考虑在香港上市，计划筹集大约5亿美元 (约39亿港元)。报道引述知情人士称，阶跃星辰已与潜在顾问就股权出售事宜进行磋商，最早可能在今 年上市，其发行规模和上市时间等细节可能有所变化。 ...

别再只盯着 "新三样"，中国正在用一种看不见的商品，收割全球 AI 市场。OpenRouter 最新周度数据击 穿认知：全球前十模型总 Token 量 8.7 万亿，中国模型独占 5.3 万亿，市占率高达 61%。MiniMax M2.5 以 2.45 万亿登顶榜首，Kimi K2.5、智谱 GLM-5、DeepSeek V3.2 全线霸榜 —— 这不是偶然的技术突 围，是中国 AI 从 "跟跑" 到 "主导" 的历史性拐点。 2. 字节生态：弹性算力之王 一、叙事核心：Token = 数字电力，一场永不枯竭的出口革命 Token 不是代码，是AI 时代的通用货币，更是最纯粹的电力衍生品。一个美国开发者调用智谱 API，数 据跨洋抵达中国数据中心，GPU 燃烧中国电力完成推理，结果再传回北美。电力从未出境，价值却通 过 Token 完成跨境交付，成本结构里电力 + 算力占比超 70%。中国低成本绿电 + 规模化算力集群，正 把能源优势，转化为全球 AI 服务的定价权。 这是比传统出口更性感的商业模式：不用集装箱、不耗物流、零库存，一度电变成百倍价值的智能服 务，源源不断流向全球。当全球还在比拼硬件制造，中国已 ...

Core Insights - Shanghai-based AI startup Jiyue Xingchen is considering an IPO on the Hong Kong Stock Exchange, aiming to raise approximately $500 million (around 3.4 billion RMB) [1] - The company has developed a series of foundational models known as the Step series, recognized for their rapid development and strong multimodal capabilities [1][2] - Jiyue Xingchen has recently completed a B+ round of financing exceeding 5 billion RMB, marking the highest single financing record in the Chinese large model sector over the past 12 months [2] Company Overview - Jiyue Xingchen was founded in April 2023 and is led by a team of industry experts, including CEO Jiang Daxin, a former Microsoft global vice president, and chief scientist Zhang Xiangyu, a top AI algorithm scientist [1] - The company has released three generations of foundational large models, with Step 3 achieving industry-leading inference efficiency [2] Strategic Developments - In January 2024, Jiyue Xingchen appointed Yin Qi, chairman of Qianli Technology, as its chairman to oversee strategic direction and technology development [3] - The company is focusing on the "AI + terminal" strategy, with significant investments in automotive, mobile, and IoT applications, as well as collaborations in finance, content creation, and retail [4] Market Context - Jiyue Xingchen's potential IPO follows the recent listings of other large model companies, including Zhiyu Technology and MiniMax, positioning it to become the third large model company to go public in China [4]

Ark Invest’s decision to purchase Tempus AI stock comes after the company’s stock declined. On Wednesday, Tempus AI’s shares closed at $53.71, reflecting a 7.32% drop. This decline followed the company’s quarterly earnings report, which revealed losses of four cents per share, aligning with analyst expectations. Despite meeting estimates, the stock faced pressure.Ark Invest’s trade involved ARK Innovation ETF (BATS: ARKK) and ARK Genomic Revolution ETF (BATS:ARKG) purchasing 212,825 shares worth nearly $11. ...

Anthropic 昨天点名 DeepSeek、月之暗面、MiniMax 三家中国 AI 实验室「蒸馏」Claude 模型，全网炸 锅。 对于此事件，RLHF （基于人类反馈的强化学习）领域最知名的研究者之一，《RLHF》一书的作者 Nathan Lambert 指出，这件事没有人们想象的那么严重，但也没有那么简单。 他认为，中国 AI 公司的基础设施非常好，取得了很多创新，也在攻克各种技术难题，但它们取得这样 的结果，靠的并不是「走捷径」。 在讨论蒸馏这件事之前，先看看 Lambert 的话为什么值得听。 Nathan Lambert 是 Allen AI 研究所的科学家，博士毕业于加州大学伯克利分校，师从机器人领域的著名 学者 Pieter Abbeel。他并非 RLHF 技术的发明者，但他写的《RLHF》这本开源书籍，如今是 AI 从业者 理解大模型训练流程的标准参考材料之一。 和到处都是的 AI 网红不一样，他是真正上手训练过大模型的人。 在 Anthropic 博客发出的当天，Lambert 就发布了一篇详细分析文章《蒸馏对于中国大模型到底有多重 要？》。他的核心论点，和主流媒体的解读方向截然不同， ...