在追求模型能力极限的同时，Skywork UniPic也坚持效率重要性的设计理念。Skywork UniPic以1.5B的 紧凑参数规模，在无CoT（思维链）的情况下取得了SOTA（"当前最佳水平"）分数，逼近部分较大模 型带CoT的0.88分；在DPG-Bench复杂指令生图基准上达到85.5分的行业SOTA水平。 据悉，Skywork UniPic在单一模型中深度融合图像理解、文本生成图像（T2I）与图像编辑三大核心任 务，构建了真正统一的多模态模型架构。 传统多模态统一模型多依赖VQ或VAE编码器来压缩视觉内容，虽然具备一定效果，但也存在局限性。 它们更侧重保留图像的视觉细节而非语义信息，这会在一定程度上削弱模型的图像理解能力。 为此，Skywork UniPic团队借鉴Harmon架构设计，并在表征方式上做出关键调整。采用MAR编码器作 为图像生成路径的视觉表征基础，同时引入SigLIP2作为图像理解路径的主干。 此外，Skywork UniPic完成端到端优化流程，能够实现生成、理解、编辑三大能力的协同训练和相互促 进，突破传统方法中能力权衡的技术瓶颈。这一架构设计不仅保持了自回归模型的简洁高效，更 ...

Core Viewpoint - The article discusses the emergence of the Skywork UniPic model, which integrates multi-modal capabilities in AI, showcasing its performance and potential impact on the industry [1][2][4]. Group 1: Model Features and Performance - Skywork UniPic is a 1.5 billion parameter model that achieves performance comparable to larger models, demonstrating high "performance density" and can run smoothly on consumer-grade graphics cards [10][12]. - The model excels in various tasks, including image understanding, text-to-image generation, and image editing, with notable scores in GenEval and DPG-Bench benchmarks [25][26][27]. - Skywork UniPic utilizes an autoregressive model architecture, allowing for deep integration of image generation within a multi-modal framework, distinguishing it from mainstream diffusion models [30][33]. Group 2: Data and Training Strategies - The model's training is based on a refined dataset approach, utilizing high-quality image-text pairs for pre-training, which enhances its semantic representation capabilities [37][42]. - A progressive multi-task training strategy is employed, focusing on one task at a time to ensure stability and performance across understanding, generation, and editing tasks [53][60]. - The team implemented specialized reward models to ensure high-quality training data, significantly improving the model's performance in both image generation and editing tasks [48][50]. Group 3: Industry Implications and Trends - The rise of native multi-modal unified models like Skywork UniPic indicates a shift in the AI landscape, emphasizing efficiency and user experience over sheer scale [61][63]. - The open-source approach taken by companies like Kunlun Wanwei is fostering innovation and accessibility in AI technology, allowing broader participation in AI development [65][68]. - The article highlights the potential for a creative explosion in AI applications, driven by user-friendly tools that lower the barriers to entry for utilizing AI [69].

Market Overview - The Shanghai Composite Index closed at 3628.53 points, above the six-month moving average, with an increase of 0.52% [1] - The total trading volume of A-shares reached 1,102.239 billion yuan [1] Stocks Breaking Six-Month Moving Average - A total of 58 A-shares have surpassed the six-month moving average today [1] - Stocks with significant deviation rates include: - Fenglong Co., Ltd. (5.52%) - Rongke Technology (4.53%) - Keyuan Wisdom (3.66%) [1] Detailed Stock Performance - The following stocks showed notable performance: - Fenglong Co., Ltd. (10.03% increase, turnover rate 7.67%, latest price 17.34 yuan) - Rongke Technology (5.47% increase, turnover rate 6.14%, latest price 18.91 yuan) - Keyuan Wisdom (4.36% increase, turnover rate 8.52%, latest price 25.60 yuan) [1] - Other stocks with positive performance include: - Zhaoyi Innovation (4.47% increase) - Yuntian Lifa (5.06% increase) [1] Additional Stocks with Minor Deviations - Stocks with smaller deviation rates that just crossed the six-month line include: - Chahua Co., Ltd. - Tangshan Port - China Gold [1]

GPT-4o的迅速走红，标注着人工智能领域多模态统一预训练模型的成熟。据了解，Skywork UniPic 延 续了GPT-4o的自回归范式，在单一模型中深度融合图像理解、文本生成图像（T2I）与图像编辑三大核 心任务，构建了真正统一的多模态模型架构。 传统多模态统一模型多依赖VQ或VAE编码器来压缩视觉内容，虽然具备一定效果，但也存在局限性， 它们更侧重保留图像的视觉细节而非语义信息，这会在一定程度上削弱模型的图像理解能力。为此， Skywork UniPic团队借鉴Harmon架构设计，并在表征方式上做出关键调整，采用MAR编码器作为图像 生成路径的视觉表征基础，同时引入SigLIP2作为图像理解路径的主干。 此外，Skywork-UniPic完成端到端优化流程，能够实现生成、理解、编辑三大能力的协同训练和相互促 进，突破传统方法中能力权衡的技术瓶颈。 7月30日，昆仑万维（300418）正式推出并开源采用自回归路线的"多模态统一预训练模型Skywork UniPic"，在单一模型中深度融合图像理解、文本到图像生成、图像编辑三大核心能力。该模型基于大 规模高质量数据进行端到端预训练，具备良好的通用性与可迁 ...

Core Viewpoint - Kunlun Wanwei (300418) officially launched and open-sourced the autoregressive "multimodal unified pre-training model Skywork UniPic" on July 30, integrating three core capabilities: image understanding, text-to-image generation, and image editing [1] Group 1 - The model is based on large-scale high-quality data for end-to-end pre-training, demonstrating strong generalization and transferability [1]

Core Insights - The music large model differs from language models in that it requires a high level of human aesthetic judgment, necessitating collaboration with professional musicians for training and optimization [1] - The company aims to achieve "music equity" by enabling users to easily create songs through an app, significantly reducing costs and production time [2] - The global music large model market is projected to reach $18.7 billion by 2025, with China accounting for approximately 32% of this market [2] Company Overview - The company, established in July 2023, has launched two applications: a one-stop music creation platform "Yinchao" and an AI-native content creation and sharing platform "Agent PI" [1] - The business model includes providing API services to B-end clients, with users able to listen to songs for free and earn revenue through community engagement [2][3] Market Context - The AI music generation sector has gained significant attention, with numerous players entering the market, including major companies like Tencent Music and ByteDance [3] - Innovative copyright and incentive mechanisms are being implemented to ensure that the core revenue from music works belongs to the creators, enhancing user engagement [3]

Group 1 - The article presents the market capitalization rankings of the top 50 Chinese technology and internet companies as of July 29, 2025 [1] - Alibaba leads the list with a market capitalization of 2,913.7 billion [3] - Xiaomi and Pinduoduo follow, with market capitalizations of 1,823.48 billion and 1,657.58 billion respectively [3] Group 2 - Meituan ranks sixth with a market capitalization of 990.12 billion [3] - Semiconductor Manufacturing International Corporation (SMIC) is in eighth place with a market cap of 530.08 billion [4] - JD.com and Kuaishou rank tenth and eleventh, with market capitalizations of 478.87 billion and 388.76 billion respectively [4] Group 3 - The list includes various companies from different sectors, such as Baidu at 307.33 billion and NIO at 109.38 billion [4][5] - The rankings reflect the competitive landscape of the Chinese tech industry, showcasing the significant market presence of these companies [1] - The data is calculated based on the market capitalization in USD, converted using the day's exchange rate [6]

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies. Core Insights - The demand for Backup Battery Units (BBUs) is experiencing a significant increase due to the rising power consumption of AI servers, with key Taiwanese companies expected to benefit from this trend [8][9][10]. - Huawei's Ascend CloudMatrix 384 SuperPod made its debut at the WAIC 2025, showcasing advancements in intelligent computing alongside other domestic competitors [11][12]. - The competition in the AI computing power sector is intensifying, with OpenAI planning to deploy 1 million GPUs by the end of the year, while Elon Musk's xAI aims for 50 million chips in five years [13][16]. Summary by Sections AI and BBU Demand - AI servers are rapidly increasing in power consumption, leading to a "straight-line upward" explosion in BBU demand. Companies like Simplo Technology, Delta Electronics, AES, and Lite-On Technology are positioned to benefit [8][9]. - The proportion of BBU modules used in ASIC racks is also increasing, and the trend towards High-Voltage DC (HVDC) technology is expected to further boost BBU demand [9][10]. Huawei and Domestic Competitors - At WAIC 2025, Huawei's Ascend CloudMatrix 384 was highlighted, achieving the largest scale of 384-card high-speed bus interconnection. Major clients include Baidu, Meituan, and JD.com [11][12]. AI Computing Power Arms Race - OpenAI is pursuing a strategy for computing independence through self-developed chips and partnerships, with a goal to shift 75% of its computing resources to its Stargate project by 2030. AI capital expenditures are projected to reach $360 billion in 2025 [13][16]. - Meta has been actively recruiting talent from DeepMind, indicating a competitive landscape for AI expertise [14].

金十图示：2025年07月28日（周一）中国科技互联网公司市值排名TOP 50一览 | 8 | | 中芯国际 | 541.27 | | | --- | --- | --- | --- | --- | | 9 | | 东方财富 | 535.73 | | | 10 | | 京东 | 478.29 | | | II | 86 | 快手-W | 398.09 | | | 12 | | 腾讯音乐 | 329.92 | -1 + | | 13 | | 理想汽车 | 316.64 | -1 | | 14 | Baics Bar | 百度 | 312.14 | -1 + | | 15 | | 贝壳 | 232.31 | -1 4 | | 16 | 8 | 同花顺 | 221.39 | 11-1 | | 17 | | 小鹏汽车 | 180.83 | -1 | | 18 | | 中通快递 | 162.27 | -1 4 | | 19 | | 科大讯飞 | 158.89 | -1 | | 20 | | 蔚来 | 111.42 | -1 + | | 21 | | 一六零 0 | 105.26 | -1 | | 22 | | 宝信软件 ...

Investment Rating - Industry Rating: Recommended [5] Core Views - The AI industry is expected to enter a "technology + policy" resonance upward cycle [4][40] - Major overseas companies are continuously advancing their AI businesses, with Google reporting a cloud revenue of $13.624 billion in Q2 2025, a year-on-year growth of 31.67% [12][17] - The World Artificial Intelligence Conference was successfully held, indicating China's leading position in the global AI revolution and the likelihood of supportive policies being introduced [2][25] - Global large model technology is expected to accelerate iteration, with OpenAI's GPT-5 anticipated to be released soon, integrating multiple internal technologies [3][40] Summary by Sections 1. Overseas Major Companies' AI Business Progress - Google reported Q2 2025 earnings with cloud revenue reaching $13.624 billion, maintaining a growth rate of approximately 30% since 2024 [12][17] - The search engine business also showed resilience, with revenue of $54.190 billion in Q2 2025, a year-on-year increase of 11.71% [20] - The Gemini application has over 450 million monthly active users, with a significant increase in daily requests [21][23] 2. Successful Hosting of the World Artificial Intelligence Conference - The conference featured over 800 companies and 3,000 cutting-edge exhibits, highlighting China's commitment to AI development [2][25] - Premier Li Qiang emphasized the importance of inclusive AI development and international cooperation during his speech [26][40] 3. Acceleration of Global Large Model Technology Iteration - OpenAI's GPT-5 is expected to be released soon, integrating various technologies to handle text, code, images, and tool calls [3][40] - The successful launch of GPT-5 could enhance confidence in the commercial application of AI technologies [3][40] 4. Investment Recommendations - Beneficial stocks include: - Large Models: iFlytek, Kunlun Wanwei [4] - AI Programming Applications: Zhuoyi Information, Dingjie Software, Hand Information [4] - AI Office Applications: Kingsoft Office, Foxit Software, Hehe Information [4] - AI Multi-modal: Wanjing Technology, Hongsoft Technology, Meitu Company [4] - AI Education Applications: Jiafa Education, Jingyeda [4] - AI Medical Applications: Weining Health, Jiahe Meikang, Rundata Medical [4]