文章目录AI科技热点日报 | 2026年07月01日 今日摘要一、Anthropic推出Claude ScienceAI正式进入科研流水线事件概要来源 / Sources二、谷歌发布Nano Banana 2 Lite4秒出图、每千张仅收费0.034美元事件概要来源 / Sources三、美团发布LongCat-2.0万亿参数大模型业界首个国产算力全流程训练事件概要来源 / Sources四、八部门联合发文工业互联网核心产业增加值目标2.5万亿事件概要来源 / Sources五、优必选U1系列全尺寸人形机器人订单破万9月16日启动交付事件概要来源 / Sources六、深圳兔展智能自研UniWorld视觉大模型跻身国际第一阵营事件概要来源 / Sources行业数据速览媒体人解读参考来源英文来源中文来源 相关阅读AI科技热点日报 | 2026年07月01日综合整理自The Verge、TechCrunch、Wired、MIT Technology Review、Reuters Technology、VentureBeat、钛媒体、36氪、虎嗅、新华网科技、网易科技、雷锋网等国内外主流科技媒体当日报道聚焦人工智能领域核心动态附资深媒体人综合解读。 今日摘要2026年7月1日AI产业在模型开源、科研工具、产业政策、硬件商业化四条战线上同步推进。在模型侧美团于6月30日发布并宣布即将开源LongCat-2.0这是业界首个在五万卡国产算力集群上完成全流程训练与推理的万亿参数大模型总参数规模1.6万亿预训练数据覆盖超过30T tokens。在科研工具侧Anthropic于6月30日正式推出Claude Science深度整合60余个科学数据库与跨代理工作流正式吹响AI进入科研流水线的号角同一天谷歌发布Nano Banana 2 Lite图像生成模型和Gemini Omni Flash视频生成模型4秒出图、每千张图像仅收费0.034美元再次刷新性价比记录。在政策侧工信部等八部门于6月30日联合印发《关于推动工业互联网高质量发展的实施意见》目标到2030年工业互联网核心产业增加值突破2.5万亿元。在硬件商业化侧优必选于6月30日发布超仿生人形机器人U1系列全渠道订单已突破13361台将于9月16日开始交付存储芯片领域闪迪上半年股价累计涨幅超过850%HBM与高带宽闪存需求持续井喷。On July 1, 2026, the AI industry advanced simultaneously across four battlegrounds: open-source model releases, scientific-research tools, industrial policy and hardware commercialization. On the model front, Meituan on June 30 released and announced the upcoming open-sourcing of LongCat-2.0, the industry’s first trillion-parameter large model trained and inferred entirely on a 50,000-card domestic computing cluster, with a total parameter scale of 1.6 trillion and pretraining data covering over 30 trillion tokens. In scientific-research tools, Anthropic on June 30 officially launched Claude Science, deeply integrating over 60 scientific databases and cross-agent workflows, formally sounding the trumpet for AI entering the scientific-research pipeline. On the same day, Google released the Nano Banana 2 Lite image generation model and Gemini Omni Flash video generation model, generating an image in just 4 seconds at a cost of only $0.034 per 1,000 images, once again setting a new性价比 record. On the policy front, the Ministry of Industry and Information Technology and seven other departments jointly issued the “Implementation Opinions on Promoting High-Quality Development of Industrial Internet” on June 30, setting a target of breakthroughs in core industrial internet added value exceeding 2.5 trillion yuan by 2030. On the hardware commercialization front, UBTECH on June 30 released its ultra-biomimetic humanoid robot U1 series, with full-channel orders exceeding 13,361 units and delivery scheduled to begin on September 16. In the storage-chip sector, SanDisk’s share price surged over 850 percent in the first half of the year, with demand for HBM and high-bandwidth flash memory continuing to surge.一、Anthropic推出Claude ScienceAI正式进入科研流水线事件概要Anthropic于6月30日正式发布Claude Science这是一款专为科研人员打造的AI工作台旨在将分散的科研工具整合至统一环境让AI代理端到端运行分析并配备专门的审核代理检查引文和计算结果。这一产品的发布标志着Anthropic在科研领域完成了迄今最深度的一次布局也是其冲刺IPO前争夺企业付费客户的重要商业动作。Claude Science基于现有Claude模型构建核心协调代理可调用超过60个预置技能和连接器覆盖基因组学、蛋白质组学、结构生物学和化学信息学等领域并能生成其他专业代理。独立的审核代理负责实时检查引文和计算标记并纠正错误以应对AI模型捏造引文和数据的常见问题。该软件首先以测试版形式向企业付费用户开放可在macOS或Linux本地运行也可通过SSH或HPC登录在远程服务器上运行大型作业由代理规划后提交至实验室自有集群或按需计算账户工作负载可从单GPU扩展至数百个数据无需离开实验室环境即可完成处理。Anthropic officially released Claude Science on June 30, an AI workbench specifically designed for researchers, aiming to integrate scattered scientific tools into a unified environment, enabling end-to-end AI agent analysis with a dedicated review agent checking citations and computational results. The launch marks Anthropic’s deepest foray into the scientific-research domain to date and is a significant commercial move to win enterprise paid customers ahead of its IPO. Claude Science is built on existing Claude models; its core coordination agent can call over 60 preset skills and connectors covering genomics, proteomics, structural biology and cheminformatics, and can also generate other specialized agents. An independent review agent is responsible for real-time checking of citations and computations, flagging and correcting errors to address the common problem of AI models fabricating citations and data. The software will initially be available in beta to enterprise paid users, running locally on macOS or Linux or via SSH or HPC login on remote servers, with large jobs planned by the agent and submitted to the lab’s own clusters or on-demand computing accounts, scaling from a single GPU to hundreds, while data never needs to leave the laboratory environment.来源 / SourcesAnthropic推出Claude Science AI科研工作台 - 腾讯新闻Anthropic推出Claude Science:AI开始接管科研流水线 - 腾讯快报Claude Science官方发布 - Anthropic7月1日外盘头条:闪迪上半年涨超850% 谷歌发布全新AI创作工具 Anthropic推出Claude Science - 环球市场播报二、谷歌发布Nano Banana 2 Lite4秒出图、每千张仅收费0.034美元事件概要当地时间6月30日谷歌发布了其最新的两款人工智能创作工具Nano Banana 2 Lite图像生成模型和Gemini Omni Flash视频生成模型。Nano Banana 2 Lite可在4秒内生成一张图像延迟较此前版本明显下降适合需要快速反复修改方案或在短时间内批量生成大量图像的工作场景。生成一张1K分辨率图像约需0.0336美元约合0.23元人民币每生成1000张图像仅收费0.034美元相比标准版Nano Banana成本大幅降低。Nano Banana 2 Lite现已登陆谷歌AI Studio、Gemini API和Gemini Enterprise Agent Platform并将取代初代Nano Banana。Gemini Omni Flash则针对高质量视频生成和对话式编辑进行了优化与Nano Banana 2 Lite共同构成谷歌面向创作者和企业的完整AI创作工具矩阵。On June 30, Google released two new AI creation tools: the Nano Banana 2 Lite image generation model and the Gemini Omni Flash video generation model. Nano Banana 2 Lite can generate an image in just 4 seconds, with significantly reduced latency compared to previous versions, making it ideal for scenarios requiring rapid iterative modifications or batch generation of large volumes of images in a short time. Generating a 1K-resolution image costs approximately $0.0336 (about RMB 0.23), with pricing of only $0.034 per 1,000 images generated, substantially lowering costs compared to the standard Nano Banana. Nano Banana 2 Lite is now available on Google AI Studio, the Gemini API and the Gemini Enterprise Agent Platform, and will replace the original Nano Banana. Gemini Omni Flash is optimized for high-quality video generation and conversational editing, and together with Nano Banana 2 Lite forms Google’s complete AI creation tool matrix for creators and enterprises.来源 / Sources谷歌发布两款AI新模型 低成本出图视频编辑无缝衔接 - 腾讯快报谷歌推出AI生图模型Nano Banana 2 Lite:4秒出图,1元钱做四张 - 钛媒体谷歌推出AI生图模型 Nano Banana 2 Lite - 腾讯快报7月1日外盘头条 - 环球市场播报三、美团发布LongCat-2.0万亿参数大模型业界首个国产算力全流程训练事件概要6月30日美团正式发布新一代基础大模型LongCat-2.0并宣布将于近期开源Infra框架、推理引擎、模型参数等核心技术。这是业界首个在五万卡国产算力集群上完成全流程训练与推理的万亿参数大模型总参数规模1.6万亿平均激活约480亿动态范围330亿至560亿MoE架构设计。LongCat-2.0预训练数据规模超过30T tokens覆盖中文、英文、多语言和代码等多类数据原生支持1M超长上下文可一次处理百万字级输入。其架构设计围绕在真实Agentic Coding任务中更高效、更稳定地完成代码理解、生成与执行这一核心目标深度适配Claude Code、OpenClaw、Hermes等主流Harness测试框架。LongCat-2.0预览版于今年4月底以匿名方式接入全球最大大模型API路由平台OpenRouter截至6月底其总调用量已跻身全球前三性能接近Claude Opus 4.6。优必选CTO评价称该模型的发布意味着国产AI基础设施从可用跨越至好用的临界点。On June 30, Meituan officially released its next-generation foundation model LongCat-2.0 and announced it would open-source its Infra framework, inference engine and model parameters in the near term. This is the industry’s first trillion-parameter large model with the entire training and inference process completed on a 50,000-card domestic computing cluster, with a total parameter scale of 1.6 trillion, averaging approximately 480 billion activated parameters, a dynamic range of 33 billion to 56 billion, and an MoE architecture design. LongCat-2.0’s pretraining data scale exceeds 30 trillion tokens, covering Chinese, English, multilingual and code data, natively supporting a 1M long context and processing up to one million characters in a single pass. Its architectural design centers on the core goal of “more efficiently and stably completing code comprehension, generation and execution in real Agentic Coding tasks,” deeply adapting to mainstream harnesses including Claude Code, OpenClaw and Hermes. The LongCat-2.0 preview version was connected anonymously to OpenRouter, the world’s largest LLM API routing platform, at the end of April this year, and by the end of June its total call volume had climbed into the global top three, with performance close to Claude Opus 4.6. UBTECH’s CTO commented that the release signals domestic AI infrastructure crossing from “usable” to “reliable.”来源 / Sources美团发布LongCat-2.0大模型,其测试版本位列OpenRouter全球总调用量前三 - 腾讯快报美团正式发布新一代万亿参数大模型LongCat-2.0 - 财联社/腾讯新闻美团发布并开源新一代万亿参数大模型LongCat-2.0 - 腾讯新闻美团开源万亿参数大模型LongCat-2.0 - 腾讯新闻四、八部门联合发文工业互联网核心产业增加值目标2.5万亿事件概要6月30日在2026工业互联网大会上工业和信息化部、国家发展改革委、人力资源社会保障部、应急管理部、中国人民银行、国务院国资委、市场监管总局、中国证监会等八部门联合印发《关于推动工业互联网高质量发展的实施意见》。《实施意见》提出两个阶段发展目标到2030年工业互联网核心产业增加值突破2.5万亿元新型基础设施实现规模化部署建设5万张工业5G专网打造5个左右具有国际影响力的综合型平台融合应用实现207个工业中类全覆盖重点行业规上工业企业安全分类分级普及率达到80%到2035年建成全球领先的工业互联网基础设施体系和国际先进的技术产业体系。在夯实融合基础设施底座方面八部门提出推动工业互联网基础设施和智算设施、超算设施等算力基础设施一体规划、同步建设探索构建工业算力网络体系强化算力互联互通增强对海量异构数据高速处理和深度加工能力深度赋能工业领域大模型训练、工业元宇宙实时交互等场景。On June 30, at the 2026 Industrial Internet Conference, eight departments—the Ministry of Industry and Information Technology, the National Development and Reform Commission, the Ministry of Human Resources and Social Security, the Ministry of Emergency Management, the People’s Bank of China, the State-owned Assets Supervision and Administration Commission, the State Administration for Market Regulation and the China Securities Regulatory Commission—jointly issued the “Implementation Opinions on Promoting High-Quality Development of Industrial Internet.” The opinions set two-stage development goals: by 2030, core industrial internet added value is to exceed 2.5 trillion yuan, with new-type infrastructure achieving large-scale deployment, 50,000 industrial 5G private networks built, about 5 internationally influential comprehensive platforms created, integrated applications covering all 207 industrial sub-sectors, and key industries’ above-scale enterprise safety classification rates reaching 80 percent; by 2035, the goal is to build a globally leading industrial internet infrastructure system and internationally advanced technology and industrial system. On consolidating the integrated infrastructure base, the eight departments proposed integrated planning and synchronized construction of industrial internet infrastructure alongside intelligent computing and supercomputing facilities, exploring the construction of an industrial computing network system, strengthening computing interconnection, enhancing high-speed processing and deep processing of massive heterogeneous data, and deeply empowering industrial large-model training and real-time industrial metaverse interaction scenarios.来源 / Sources八部门联合发文 擘画工业互联网2.5万亿产业新图景 - 腾讯快报推动工业互联网高质量发展 - 新华网科技八部门联合发文,工业互联网利好政策来了 - 腾讯快报工信部等八部门印发《关于推动工业互联网高质量发展的实施意见》 - 人民网/新华网五、优必选U1系列全尺寸人形机器人订单破万9月16日启动交付事件概要6月30日“人形机器人第一股优必选在深圳举办2026年度全球发布会正式推出旗下消费级品牌优世界”UWORLD的首款产品——全尺寸超仿生人形机器人U1系列。U1系列分为U1 Lite、U1 Pro和U1 Ultra三个版本其中U1 Lite售价11.98万元、U1 Pro售价16.98万元、U1 Ultra男版售价99万元、女版售价88万元。截至当日下午3时该系列机器人线上线下全渠道订单量已超过13361台成为目前全球唯一具备规模化量产能力的全尺寸超仿生人形机器人产品。优必选首席品牌官谭旻表示U1系列已于6月2日起量产将于9月16日开始交付力争在年底前完成全部订单交付。这一数字对比2025年全年优必选全尺寸具身智能人形机器人仅1079台的销量实现了十倍以上的单款新品订单规模增长。发布会上优必选还宣布成立人工智能与机器人科技伦理委员会并同步启动100台定制仿生机器人人机伴独工程扶持计划。On June 30, “humanoid robot first stock” UBTECH held its 2026 Global Launch Event in Shenzhen, officially launching the first product of its consumer brand “UWORLD”—the full-size ultra-biomimetic humanoid robot U1 series. The U1 series includes three versions: U1 Lite priced at 119,800 yuan, U1 Pro at 169,800 yuan and U1 Ultra at 990,000 yuan for the male version and 880,000 yuan for the female version. As of 3 p.m. that day, the robot series had surpassed 13,361 units in full-channel orders across online and offline channels, making it the currently only full-size ultra-biomimetic humanoid robot product with规模化量产 capability globally. UBTECH chief brand officer Tan Min said the U1 series entered mass production on June 2 and will begin delivery on September 16, striving to complete all order deliveries by year-end. This number represents more than ten times the single-product new order scale compared to UBTECH’s 2025 full-size embodied intelligence humanoid robot sales of only 1,079 units for the entire year. At the launch event, UBTECH also announced the establishment of an AI and Robotics Technology Ethics Committee, and simultaneously launched a “human-robot companionship project” support plan offering 100 customized biomimetic robots.来源 / Sources优必选发布全球首款全尺寸超仿生机器人U1系列,订单已超1.3万台 - 腾讯快报11.98万起!优必选发布首款超仿生人形机器人! - 腾讯快报以情感大模型重新定义人形机器人家庭场景,优必选超仿生机器人首发订单破万 - 腾讯快报优必选超仿生人形机器人订单破万:50余款亮相,年交付目标1万台 - 腾讯快报六、深圳兔展智能自研UniWorld视觉大模型跻身国际第一阵营事件概要深圳兔展智能科技有限公司在AI视觉领域持续深耕其自研的UniWorld系列视觉大模型是国内最早实现理解、生成、编辑统一架构的视觉大模型真正做到了结构、语义与计算的深度协同——同一个模型既能看懂图像也能生成图像还能在理解的基础上生成更准确的图像。早在2023年多模态技术刚刚起步之时兔展智能便着手探索理解与生成的统一UniWorld V1版本早于谷歌的Nano Banana三个月推出并同步开源。2024年兔展智能联合北京大学深圳研究生院推出Open-SoraPlan开源项目在GitHub上线一周即获得超过5000名开发者标星成为全球技术社群引用率最高的视觉开源模型之一。2025年兔展智能推出全球首款实现三维一致性生成的AI产品AnyReal同年其自研多模态大模型兔灵及中文艺术海报生成技术获国家专利。目前兔展智能已获评国家级专精特新小巨人企业2025年研发投入超5亿元人民币服务超过4100万企业客户包括华为、荣耀等知名企业并已完成股份制改革筹备上市。Shenzhen Rabbit Pre-Intelligence Technology Co. continues to deepen its presence in the AI vision field, with its self-developed UniWorld series of visual large models being the earliest domestic visual models to achieve a unified architecture of understanding, generation and editing—truly achieving deep collaboration among structure, semantics and computation, where the same model can comprehend images, generate images, and produce more accurate images based on comprehension. As early as 2023, when multimodal technology was just emerging, Rabbit Pre-Intelligence began exploring the unification of understanding and generation. The UniWorld V1 version was launched three months before Google’s Nano Banana and was open-sourced simultaneously. In 2024, Rabbit Pre-Intelligence, together with the Peking University Shenzhen Graduate School, launched the Open-SoraPlan open-source project, which received over 5,000 developer stars on GitHub within one week of launch, becoming one of the most-cited visual open-source models in the global technical community. In 2025, Rabbit Pre-Intelligence launched AnyReal, the world’s first AI product achieving three-dimensional consistency in generation, and the same year its self-developed multimodal large model “TuLing” and Chinese art poster generation technology received national patents. Currently, Rabbit Pre-Intelligence has been recognized as a national-level specialized, refined, and new “little giant” enterprise, with RD investment exceeding 500 million yuan in 2025, serving over 41 million enterprise customers including well-known companies such as Huawei and Honor, and has completed shareholding system reform preparing for listing.来源 / SourcesAI赋能,一句话办成一件事:深圳兔展智能以自研视觉大模型,助力广东增强人工智能产业竞争力 - 腾讯新闻深圳兔展智能科技有限公司经营信息 - 企查查AI视觉先锋兔展智能:十年创新引领国产大模型突围 - 搜狐兔展智能如何利用视觉大模型实现AI技术的突破与创新 - 搜狐行业数据速览指标数据备注美团LongCat-2.0总参数1.6万亿首个国产算力全流程训练的万亿参数模型LongCat-2.0预览版OpenRouter排名全球前三性能接近Claude Opus 4.6Claude Science预置科学数据库60个覆盖基因组学、蛋白质组学、结构生物学、化学信息学Nano Banana 2 Lite出图速度4秒/张每千张图像仅收费0.034美元八部门工业互联网目标2030核心产业增加值2.5万亿元建设5万张工业5G专网优必选U1系列订单量13,361台9月16日开始交付力争年底前完成U1系列2025年对标销量1,079台U1系列单款新品订单规模为去年近10倍兔展智能2025年研发投入超5亿元国家级专精特新小巨人企业闪迪2026上半年股价涨幅超850%HBM与高带宽闪存需求井喷媒体人解读2026年7月1日这一天AI产业呈现出两条清晰的主线一条是模型平权——美团LongCat-2.0用国产五万卡集群跑出全球前三的OpenRouter调用量、Anthropic Claude Science将AI科研工具从实验室普惠到企业付费用户、谷歌Nano Banana 2 Lite以每千张图像0.034美元的极致成本将AI创作门槛压至几乎为零——三条路线共同指向同一个趋势大模型正在从少数公司的军备竞赛演变为全社会的基础设施。另一条主线是硬件商业化元年——优必选U1系列万台订单标志着人形机器人在消费级市场的破冰八部门工业互联网政策为AI与制造业融合提供了制度保障而存储芯片领域闪迪股价半年翻8.5倍则表明上游算力需求仍未见顶。两条主线相互缠绕模型平权降低了硬件的应用门槛硬件商业化则为模型迭代提供了真实场景和资金回报。展望下半年AI产业的最大变量将不再是模型有多强而是模型和硬件在多少场景里真正用起来了。On this day, July 1, 2026, the AI industry presents two clear main threads: one is “model democratization”—Meituan’s LongCat-2.0 achieving top-three OpenRouter usage with a domestic 50,000-card cluster, Anthropic’s Claude Science bringing AI research tools from the lab to enterprise paid users, and Google’s Nano Banana 2 Lite pushing AI creation barriers down to near-zero with an极致 cost of $0.034 per 1,000 images—all three routes converging toward the same trend: large models are evolving from “arms races among a few companies” to “social infrastructure for all.” The other main thread is “hardware commercialization year one”—UBTECH’s U1 series with 10,000-plus orders marking the breakthrough of humanoid robots in the consumer market, the eight-department industrial internet policy providing institutional guarantee for AI and manufacturing integration, and SanDisk’s share price surging 8.5 times in half a year indicating upstream computing demand has not yet peaked. These two threads intertwine: model democratization lowers the application threshold for hardware, while hardware commercialization provides real-world scenarios and financial returns for model iteration. Looking ahead to the second half of the year, the biggest variable in the AI industry will no longer be “how powerful the models are,” but “in how many scenarios are models and hardware truly being used.”参考来源英文来源Claude Science官方发布 - AnthropicSK hynix Ships Samples of 12-Layer Next-Gen ‘HBM4E’ - SK hynix NewsroomGoogle Nano Banana 2 Lite - Google AI Blog中文来源Anthropic推出Claude Science - 腾讯新闻美团发布LongCat-2.0万亿参数大模型 - 腾讯新闻八部门联合发文 工业互联网2.5万亿产业新图景 - 腾讯快报优必选U1系列订单超1.3万台 - 腾讯快报深圳兔展智能视觉大模型 - 腾讯新闻谷歌发布Nano Banana 2 Lite - 钛媒体/腾讯快报7月1日外盘头条 - 环球市场播报本文内容综合整理自上述公开信息来源媒体人解读仅代表个人观点仅供参考。发布日期2026年07月01日 相关阅读D60: 给读者的信你的下一步是什么开发中常用素材格式完整清单D47 《数据治理从混乱到可控》防护体系的「治理」篇一人公司OPC典型案例与商业模式研究报告主流 Claw 类产品技术与商业差异全景报告常见误区管理者最容易踩的 5 个坑团队 AI 成熟度自评模型
