Core Viewpoint - The article discusses the launch of Rakuten AI 3.0, Japan's purportedly strongest AI model, which has been revealed to be a Japanese version of the Chinese open-source model DeepSeek V3, raising concerns about originality and transparency in Japan's AI development [4][8][39]. Group 1: Model Launch and Claims - Rakuten AI 3.0 was launched with claims of being Japan's largest and most powerful AI model, supported by the Japanese Ministry of Economy, Trade and Industry through the GENIAC project [6][12]. - The model boasts an impressive 700 billion parameters, placing it among the top tier of current open-source models [6][12]. - Despite the initial excitement, it was quickly discovered that Rakuten AI 3.0 is essentially a Japanese "shell" version of DeepSeek V3, leading to disappointment among Japanese users [8][11]. Group 2: Community Response and Transparency Issues - The open-source community quickly identified that Rakuten AI 3.0's architecture is based on DeepSeek V3, which was not disclosed during its launch [11][13]. - The lack of acknowledgment of DeepSeek V3 in the model's release raised significant concerns about adherence to open-source principles, particularly regarding the retention of original licenses and attributions [14][25]. - Japanese users expressed frustration, questioning the use of taxpayer money for a model that essentially repackages an existing Chinese product [14][25]. Group 3: Performance Metrics - Rakuten AI 3.0 has shown competitive performance metrics, surpassing GPT-4o in various Japanese language benchmarks, indicating its capabilities in local contexts [16][24]. - The model's cost efficiency is notable, operating at approximately 10% of the cost of leading closed-source models due to its MoE sparse architecture [23][24]. - However, the performance advantages are largely reflective of the underlying DeepSeek V3 architecture, raising questions about the innovation claims made by Rakuten [25][39]. Group 4: Broader Context of Japan's AI Industry - The article highlights a broader trend in Japan's AI industry, where many models are built on existing foreign architectures, indicating a reliance on external technologies rather than homegrown innovation [27][39]. - Japan's IT sector has faced stagnation over the past three decades, struggling to keep pace with global advancements in AI and technology [31][39]. - The situation with Rakuten AI 3.0 exemplifies the challenges faced by Japanese companies in establishing a strong presence in the rapidly evolving AI landscape, often resorting to leveraging existing models rather than developing original solutions [38][39].

3.吉利副总裁回应"董明珠购车"：其个人订购3辆极氪009，连同下属员工等合计订购9辆 4.百度推动大模型与搜推业务融合，何径舟任百度App研发中心组织负责人 5.真安全、真绿色、真生态，2026奇瑞汽车电池之夜发布能源新战略 6.字节内部发布"龙虾"相关安全规范，面向员工推出ByteClaw 要闻提示 NEWS REMIND 1. 「日本最强AI」塌房了！扒开代码全是DeepSeek，日本网友集体破防 2.腾讯年报披露：人均年薪成本超百万 7.三星电子工会拟5月全面停工18天！93%员工赞成：不涨工资就不干了 8.卓驭科技计划港股上市，Pre-IPO轮融资20亿 今日头条 HEADLINE NEWS 「日本最强AI」塌房了！扒开代码全是DeepSeek，日本网友集体破防 3月18日消息，据媒体报道，日本乐天集团在17日刚高调发布号称"日本国内最大规模"的高性能AI模型 Rakuten AI 3.0，结果不到12小时就被技术社区扒得底朝天，直接实锤核心架构全抄中国深度求索的开源 模型DeepSeek V3，初期还完全没标注原创方，引爆日本全网舆论。 报道称，网友直接扒出乐天当天上传到Hugging Face开 ...

核心参数、层数、维度与中国 DeepSeek V3 完全一致，乐天只做了少量日文数据微调。 很离谱的是， 乐天最初上传时 直接删除了 DeepSeek 开源许可证 ，被抓包后才匆忙补上一个 NOTICE 文件。 官方通稿、发布会只说用上了社区开源成果， 全程不提"DeepSeek"， 刻意包装成"日本自研"，涉嫌虚假宣 传。 目前日本网友集体炸锅，批评其拿纳税人补贴搞"换皮套壳"，违背开源精神，"欺骗国民感情"。 2026 年 3 月 17 日，日本乐天高调发布号称"日本最大规模"的 AI 模型 Rakuten AI 3.0，宣称参数约 7000 亿、 日语性能超越 GPT-4o，还拿到日本政府 GENIAC 项目补贴。 不过它发布不到 12 小时，就被全球开发者彻底扒穿。 有开发者在 HuggingFace 平台查看模型配置文件发现， config.json 里明明白白写着：架构 = DeepseekV3ForCausalLM 。 DeepSeek V3 采用 MIT 开源协 议 （ 这 几乎没有限制性条款，主要要求是在分发时保留原始的版权声明和许可声明 ） ，其他厂商 可以合规用，但隐瞒来源就是信用抄袭。 ...