开源大模型
Search documents
中国信通院:超一半金融企业积极规划内部开源的协作机制
Zhong Guo Qing Nian Bao· 2025-07-24 10:04
Group 1 - The "support for the development of open-source technology communities" goal outlined in the "14th Five-Year Plan" is being actively implemented across various sectors of the financial industry [1] - Over 58% of the financial sector is actively planning internal open-source collaboration mechanisms to enhance deep collaboration and sharing among technology teams, accelerating the industry's transition to intelligence and platformization [1] - The China Academy of Information and Communications Technology (CAICT) has facilitated open-source governance assessments for numerous financial institutions, including Agricultural Bank of China, Industrial and Commercial Bank of China, and China Construction Bank, creating replicable and scalable standardized practices [1] Group 2 - More than 50 financial enterprises have collaboratively established an innovation platform for technological collaboration and achievement transformation within the financial open-source community [1] - Open-source models are recognized as a new production method in the digital age, effectively reducing costs associated with technological innovation, resource allocation, and industrial transformation [1] - According to CAICT's research, the application rate of the DeepSeek series open-source models in financial enterprises is as high as 100%, while the Tongyi Qianwen series exceeds 70%, indicating that open-source is becoming the mainstream model for digital technology innovation [1] Group 3 - Representatives from China Ping An Life, Agricultural Bank of China, CAICT, China Construction Bank, and Industrial and Commercial Bank of China jointly released a roadmap for the construction of the financial open-source system and initiated a collection of excellent case studies [2] - A pilot program for assessing the innovation and development capabilities of financial open-source initiatives has also been launched [2]
Qwen3小升级即SOTA,开源大模型王座快变中国内部赛了
量子位· 2025-07-22 04:35
Core Viewpoint - The article discusses the rapid advancements in open-source large models in China, highlighting the release and performance of the Qwen3 model, which has shown significant improvements over its predecessor and competitors in various benchmarks [1][24]. Group 1: Model Updates and Performance - Qwen3 has been upgraded to a model with 235 billion parameters, which is only a quarter of Kimi K2's 1 trillion parameters, yet it surpasses Kimi K2 in benchmark performance [2][3]. - The new model enhances understanding of 256K long contexts and is a causal language model utilizing a Mixture of Experts (MoE) architecture [8][12]. - The model includes 94 layers, employs grouped query attention (GQA) mechanisms, and activates 8 out of 128 experts during inference [8][12]. Group 2: Benchmark Performance - In benchmark tests, Qwen3 shows improved accuracy in various categories, such as AIME25, where accuracy increased from 24.7% to 70.3%, indicating strong mathematical reasoning capabilities [13][15]. - Compared to Kimi K2 and DeepSeek-V3, Qwen3 demonstrates superior performance across multiple metrics, including instruction following, logical reasoning, and text understanding [12][15]. Group 3: Market Context and Competition - The article notes that the competitive landscape is shifting, with Qwen3 challenging Kimi K2 shortly after its release, indicating a dynamic environment in the open-source model sector [25]. - The release of Qwen3 coincides with NVIDIA's announcement of a new state-of-the-art open-source model, OpenReasoning-Nemotron, which offers various scales and local operation capabilities [17][18]. - The transition of Llama to a closed-source model and OpenAI's delay in releasing open models further emphasizes the growing importance of open-source large models in the Chinese market [24].
游戏ETF(516010)涨超1.1%,版号放量叠加新游表现提振行业信心
Mei Ri Jing Ji Xin Wen· 2025-07-21 02:17
Group 1 - The core viewpoint is that 2023 is expected to be a breakout year for closed-source general large models, while 2025 is anticipated to reshape the landscape for open-source large models in China [1] - The gaming sector is recommended for investment opportunities after market corrections, with core product high-frequency data showing continuous improvement [1] - AI applications are maturing, with the Agent landing paradigm becoming more established, leading to cost reduction and efficiency improvements in 2B sectors and enhanced experiences in 2C sectors [1] Group 2 - Commercialization processes for AI companionship and AI education are accelerating, aligning with personalized needs and high willingness to pay [1] - The IP derivative sector is experiencing increased prosperity, with accelerated progress in licensing businesses [1] - The gaming ETF (516010) tracks the animation and gaming index (930901), reflecting the overall performance of listed companies in the animation, comics, and gaming sectors [1]
长青游戏营收压舱、新游表现决定增量,聚焦游戏板块布局窗口
Mei Ri Jing Ji Xin Wen· 2025-07-21 02:12
7月21日早盘,游戏板块表现略微震荡,游戏ETF(159869)现涨幅有所收窄,涨近1%。游戏ETF(159869)已连续5个交易日获资金净流入,累 计"吸金"达15.18亿元,备受资金青睐。 华创证券指出,2023年是闭源通用大模型的爆发之年,看好2025年成为中国开源大模型爆发及应用格局重塑之年。游戏板块建议关注回调后的 布局机会,核心产品高频数据持续向好。AI应用方面,Agent落地范式逐步成熟,垂类2B降本增效及2C体验优化加速;AI陪伴、AI教育等场景 商业化进程加快,符合个性化需求且付费意愿高。IP衍生赛道景气度提升,授权业务进展加速。 据统计,2025年上半年共有812款网络游戏获得版号,涉及618家运营单位,其中国产网络游戏757款,进口网络游戏55款。从数量来看,今年 上半年游戏版号数量为近5年之最,全年有望突破1500款,接近2019年水平。从审批情况来看,今年上半年游戏版号月均过审135款,超去年同 期的115款,整体发放稳中有增。中信建投(601066)数据显示1-5月国内手游市场规模同比增长20%——在版号充足、需求大增但买量成本同 样高涨的背景下,行业呈现出长青游戏营收压舱、新游表 ...
黄仁勋评价DeepSeek和通义千问:都是世界顶尖开源大模型
Zhong Guo Zheng Quan Bao· 2025-07-17 21:03
Core Insights - The third China International Supply Chain Promotion Expo highlighted the significance of open-source AI models like DeepSeek and Tongyi Qianwen, which are considered top-tier globally, showcasing China's excellence in open-source initiatives [1][2] - NVIDIA's CEO emphasized the importance of the Chinese market for NVIDIA, describing it as one of the largest and most vibrant markets in the world [3] Group 1: AI Technology Development - AI technology has evolved from perception-based to generative AI, with significant advancements in computer vision, speech recognition, and language understanding surpassing human capabilities [1] - The future trend of AI development is expected to penetrate the physical world, leading to the rise of physical AI applications in robotics [1][2] Group 2: China's Role in AI - China leads the world in the number of AI research papers published, indicating its pivotal role in the AI technology landscape [2] - Open-source models are facilitating the formation of China's AI ecosystem and are also contributing to the development of AI ecosystems in other regions globally [2] Group 3: NVIDIA's Strategic Position - NVIDIA announced the resumption of H20 chip sales in China and the launch of a new GPU compatible with the Chinese market, signaling positive developments for the AI industry chain [3] - The company's products are being utilized in various sectors in China, including supply chain digital management and training embodied intelligent models [3] Group 4: Future Outlook - NVIDIA's technology roadmap covers nearly a decade, with the CEO indicating that there is substantial work ahead, particularly in the context of AI and chip technology advancements [3] - Innovations in silicon technology are anticipated in transistor structure, packaging technology, and silicon photonics, which will drive future developments in the chip sector [2]
K2开源大模型,会是Kimi的DeepSeek时刻吗?
Hu Xiu· 2025-07-14 03:20
Core Insights - The article discusses the emergence of MoonShot's latest open-source model K2, which has a parameter scale of 1 trillion, making it the largest open-source model currently available [2] - K2's performance in various benchmarks positions it as a strong competitor against established models like Claude 4 Opus and GPT-4.1, highlighting China's growing influence in the global AI landscape [2][4] - The competitive landscape in the AI sector is intensifying, with Chinese companies like MoonShot and MiniMax leading the charge in open-source innovation, challenging Western counterparts [4][6] Company Developments - MoonShot's K2 model has quickly gained popularity, becoming the top trending open-source model on HuggingFace shortly after its release [4] - The model's architecture incorporates fewer attention heads and more experts, enhancing efficiency in processing long contexts, which is a significant improvement over previous models [8][10] - MoonShot has disclosed a total funding amount of approximately $1.5 billion, which is significantly lower than that of its Western competitors, indicating a more efficient operational model [6] Market Impact - K2's compatibility with OpenAI and Anthropic's API formats positions it favorably in the AI application development market, potentially allowing it to capture a significant share of the market [7] - The article notes that the competitive dynamics between MoonShot and DeepSeek have intensified, with both companies releasing multiple models aimed at various AI applications [5][12] - The focus on multi-agent collaboration and the integration of various models into K2 may enhance its commercial viability and market appeal [12]
中国信通院“开源大模型+”软件创新应用典型案例入围结果公布
Huan Qiu Wang Zi Xun· 2025-07-10 03:19
Group 1 - The Global Digital Economy Conference recently held a forum to announce the shortlisted cases for the "Open Source Large Model+" software innovation applications [1] - The initiative aims to explore the application potential of open-source large models like DeepSeek across various industries, promoting technological innovation and business upgrades [4] - A total of over 100 typical cases were submitted from across the country, with 68 cases shortlisted after rigorous evaluation based on innovation, technological breakthroughs, and ecological synergy [5] Group 2 - Among the shortlisted cases, 26 were recognized as selected innovative cases due to their technical strength, innovative achievements, and application value [5] - The second round of case collection for "Open Source Large Model+" software innovation applications will begin, with results expected to be announced in the second half of 2025 [5] - The selected innovative cases will be awarded at a future event related to cloud and software security [5]
【财闻联播】柬埔寨宣布与美国达成关税协议!“网红医生”被点名,国家卫健委紧急提醒
券商中国· 2025-07-05 10:55
Macro Dynamics - The National Health Commission of China is intensifying efforts to regulate the internet health science popularization, addressing issues with "internet celebrity doctors" who misuse their authority for profit [1] - The commission emphasizes that medical quality and safety are non-negotiable, urging the public to be cautious and avoid scams [1] Market Data - The average price of pork in China's wholesale markets increased to 20.58 yuan per kilogram, up 1.7% from the previous week [3] Company Dynamics - Shanghai Lego Land officially opened on July 5, with plans for future expansion over the next 5-10 years [10] - Sales of 3C certified power banks have surged following a recall event, leading to stock shortages among some retailers [11] - Road Gang has been appointed as the chairman of Beijing Construction Group, replacing Fan Jun [12]
盘古团队声明:严格遵循开源许可证的要求
news flash· 2025-07-05 09:38
Core Viewpoint - The Pangu team emphasizes that the Pangu Pro MoE open-source model is developed and trained on the Ascend hardware platform, and it is not based on incremental training from other vendors' models [1] Group 1 - The Pangu Pro MoE open-source model includes some foundational components whose code implementations reference industry open-source practices, involving portions of open-source code from other large models [1] - The Pangu team asserts that they strictly adhere to open-source license requirements, clearly marking copyright statements for open-source code in the open-source code files [1]
盘古团队最新声明:严格遵循开源要求
第一财经· 2025-07-05 09:26
Core Viewpoint - Huawei's Noah's Ark Lab announced the release of the Pangu Pro MoE open-source model, emphasizing that it is developed and trained on the Ascend hardware platform and not based on incremental training from other vendors' models [1] Group 1 - The Pangu Pro MoE open-source model incorporates some foundational components whose code implementations reference industry open-source practices, including portions of open-source code from other large models [1] - The team stated that they strictly adhere to open-source license requirements, clearly marking copyright statements in the open-source code files [1] - This approach aligns with the common practices of the open-source community and reflects the industry's advocacy for open-source collaboration [1]