证券时报记者在现场了解到，本次东博会将AI放在了前所未有的高度上，首次设立了规模达1万平方米的AI专馆，约1200项国内外的前沿AI产品 在此亮相。来自中国的AI卫星、人形机器人、智能翻译终端和来自文莱等东盟国家的AI智慧医疗系统等产品在东博会同台亮相，汇聚成了一场跨 越山海与国界的"科技大集"。 AI穿针引线 （原标题：东博会AI含量拉满！"以智为船"扬帆四海） 开幕式上，"数字人司仪"身穿民族服装登台亮相，精准捕捉大会热词、梳理会议重点；大屏幕中，7种语言的人工智能（AI）实时同声传译打通了 中外语言的屏障……从人工办展会到人工智能办展会，9月17日召开的第二十二届中国—东盟博览会（以下简称"东博会"）实现了"AI含量"的飞跃 式提升。 "左前方行走……左转……乘电梯下一楼……"手机发出的声音，是"AI东博"智能体馆内导航的提示音。本届东博会展览面积达16万平方米，来自 60个国家约3200家企业参展，展馆众多，"迷路"成了常见现象。但扫一扫东博会上随处可见的二维码，连接"AI东博"智能体，就可以获得准确清 晰的导航。 这个AI智能体的作用不仅仅是充当地图软件那么简单，它还能充当"展品介绍员"，用手机摄像 ...

实际上，本届东博会的筹备全过程都有AI参与，这也是东博会史上头一次。记者了解到，从东博会展 馆内外主视觉设计，到吉祥物"合合""美美"的新潮穿搭，再到东博会开幕式的主题短片VCR和展会宣传 片等，全都由AI生成。东博会秘书处秘书长韦朝晖表示："从过去的人工办展办会蜕变为AI办展办会， 这是会展业的大胆尝试和重要变革。" 这种"AI震撼"，贯穿了东博会的每个角落。在开幕式上，进行演讲的嘉宾每说一句话，在巨型的LED屏 幕上，中文、英语、越南语、泰语、缅甸语、马来语、印尼语七种语言共同组成的实时翻译字幕便在屏 幕上跳动，翻译结果的精准与高效让现场嘉宾掌声雷动。 这背后，是科大讯飞提供的技术支持。科大讯飞副总裁、听见科技总经理王玮表示："在中英翻译部 分，（多语言实时字幕）采用的是已发展成熟的星火语音同传大模型，该模型已涵盖金融、医疗等垂直 领域超8万个专业词汇；针对东盟语言的翻译部分，我们将东盟语言的语法、文化语境以及专业术语等 方面进行了专项优化，同时将重要嘉宾的名字以及一些新词汇也提前加入到语料库中，以提升翻译的准 确性和适配性。" 在科大讯飞的展区，针对东盟语言进行升级的讯飞双屏翻译机2.0吸引了许多东盟参 ...

科大讯飞副总裁、听见科技总经理王玮介绍，本次东博会的多语言实时字幕服务在中英翻译部分，采用 的是星火语音同传大模型，该模型已涵盖金融、医疗等垂直领域超8万个专业词汇。 科大讯飞打造的两位数字人司仪也在开幕式上亮相。"现在马上为您捕捉本次开幕式发言中多次提及的 大会热词"数字人AA说道，大屏幕上，出现"东博会、人工智能+、中国—东盟自贸区、'一带一路'、旅 游、水果进口"等关键词云图。所有议程结束，数字人充当"信息向导"，帮助参会嘉宾快速锁定会议重 点内容。一键生成的会议纪要、新闻素材，以及通过定制化AI产品精准抓取的大会核心热词、精彩瞬 间照片，以视频、PPT两种直观形式整合呈现。 科大讯飞为此次东博会全程提供AI翻译服务之外，还将在会期发布面向东盟的星火东盟多语言大模型 底座以及讯飞翻译SaaS平台、讯飞听见多语言会议系统（东盟版）、科大讯飞中文智慧教学系统（东盟 版）等产品，并针对东盟语言升级讯飞双屏翻译机2.0。 9月17日，第22届中国—东盟博览会在广西南宁开幕。作为第22届中国—东盟博览会多语言翻译服务暨 人工智能特别合作伙伴，科大讯飞（002230）以AI助力这场国际经贸盛会实现全场景无障碍沟通。 ...

Core Insights - Tencent's Hunyuan-MT-7B model has been open-sourced after winning the international translation competition, showcasing its capabilities in translating 33 languages and understanding internet slang and gaming terminology [1][3]. Group 1: Model Performance and Features - Hunyuan-MT-7B achieved first place in the WMT2025 competition for 30 languages, outperforming larger models under strict parameter size and open-source training data requirements [3]. - The model demonstrated superior performance on the Flores200 translation evaluation dataset, leading among models of similar size and comparable to larger models [6]. - An integrated translation model, Hunyuan-MT-Chimera-7B, has also been released, which enhances translation quality by combining outputs from multiple models [6]. Group 2: Strategic Implications - The open-sourcing of Hunyuan-MT-7B indicates a strategic shift in Tencent's translation business, moving from the discontinued "Tencent Translator" to a focus on large model ecosystems [7]. - The closure of "Tencent Translator," which had previously handled 200 million translation requests daily, reflects a transition rather than a contraction of services, integrating core functionalities into the "Tencent Yuanbao" intelligent platform [7]. - Competitors in the industry, such as NetEase Youdao and iFlytek, have also launched their own large translation models, indicating a growing trend towards advanced AI translation solutions [7].

Core Viewpoint - The first half of 2025 is crucial for domestic AI computing power companies, with significant growth in shipments and revenue driven by local capabilities and the need for AI applications across various industries [3][4][12]. Group 1: Market Dynamics - The AI computing power market is experiencing a shift as domestic companies capitalize on the limitations of foreign chips, leading to rapid growth in shipments [3][4]. - Companies like DeepSeek and iFlytek are leveraging local computing power to enhance their AI models, with iFlytek's X1 model showing a 20% improvement in translation accuracy [5][6]. - The competition among enterprises is expected to accelerate AI implementation, with even a 10% efficiency or cost difference being critical for survival [12]. Group 2: Collaboration and Standards - The establishment of the "Model-Chip Ecological Innovation Alliance" aims to create a unified standard among chip manufacturers and model developers, reducing adaptation costs [9][10]. - Companies are encouraged to collaborate closely to ensure that chip designs align with the rapid iteration cycles of AI models, which currently outpace chip development timelines [9][10]. Group 3: Application Scenarios - Various sectors, including finance, education, and healthcare, are rapidly deploying AI applications, creating a diverse range of use cases that require tailored chip solutions [12][13]. - The unique ecosystem in China, similar to the mobile internet era, may provide significant opportunities for local chip companies to innovate and grow [12][13]. Group 4: Future Outlook - The AI revolution is seen as a critical juncture for Chinese chip companies, which must adapt quickly to avoid falling behind global competitors [13]. - The potential for AI to transform industries is widely recognized, and companies must leverage this technology to drive economic growth and improve operational efficiency [12][13].

Core Insights - The domestic AI computing power market is experiencing rapid growth, particularly in the first half of 2025, as local companies capitalize on opportunities created by restrictions on foreign chips [2][5][14] - Companies are actively seeking recognition for their digital intelligence platforms from major enterprises, with initial collaborations expected to yield revenue in the latter half of the year [4][12] - The development of large models in China is gaining momentum, with local firms like DeepSeek and iFlytek making significant advancements, although they still face challenges in keeping pace with international competitors like NVIDIA [6][9][14] Industry Trends - The AI application landscape is expanding across various sectors, with a consensus forming around the transformative potential of AI technologies [2][12] - The demand for AI computing power is increasing, with a notable 59% year-on-year revenue growth attributed to AI computing capabilities in one company [5] - The establishment of alliances among chip manufacturers and model developers aims to create standardized protocols, reducing adaptation costs and enhancing collaboration [10][12] Company Developments - iFlytek's Spark model, built on domestic computing power, has been upgraded to improve translation accuracy, showcasing the capabilities of local AI infrastructure [6][12] - Companies are focusing on understanding specific application scenarios to better align chip designs with market needs, indicating a shift towards more tailored solutions [13] - The competitive landscape is intensifying, with companies recognizing that even minor efficiency or cost differences can determine their survival in the market [12][14]