月之暗面发布全球最大开源AI模型Kimi K3挑战美国对手Moonshot AI Launches World's Largest Open-Source Model Kimi K3

中国人工智能初创公司月之暗面（Moonshot AI）正式发布了名为Kimi K3的全新人工智能模型。该模型被定位为目前全球参数规模最大的开源AI模型。在人工智能和机器学习领域，参数是衡量系统在训练过程中复杂程度的关键指标，通常情况下，参数数值越高，意味着模型的整体性能和处理能力越强。月之暗面此次推出这一超大规模的开源系统，其核心目标在于直接挑战美国科技巨头在人工智能领域的长期主导地位。

Chinese artificial intelligence startup Moonshot AI has officially launched its new AI model, Kimi K3, positioning it as the world's largest open-source artificial intelligence model to date. In the field of machine learning, parameters serve as a key metric to measure the complexity of AI systems during their training phase, with a higher figure generally indicating a stronger and more capable model. By introducing this massive open-source system, the Chinese developer aims to directly challenge the dominant position of leading US technology firms in the global artificial intelligence sector.

作为一家中国本土的AI先锋企业，月之暗面选择将Kimi K3进行开源，旨在为全球开发者提供强有力的技术支持。这一举措不仅展示了中国初创公司在底层大模型研发上的技术实力，也反映出国内开发者正在通过开源生态加速技术迭代，积极追赶并试图超越以OpenAI和Anthropic为代表的美国顶尖人工智能实验室，从而改变全球AI竞争的格局。

As a pioneer in China's AI landscape, Moonshot AI's decision to release Kimi K3 as an open-source model reflects a strategic effort to empower developers globally while showcasing domestic research capabilities. The launch represents a significant move by Chinese developers to narrow the technological divide, actively competing against top-tier US laboratories such as OpenAI and Anthropic by offering a highly complex, open-access alternative to proprietary American platforms.

根据月之暗面在官方博客中发布的信息，Kimi K3模型在设计之初就有着明确的技术定位，专门用于处理长程代码编写、复杂的知识工作以及高难度的推理任务。为了验证其在这些特定领域的实际效能，该公司对该模型进行了一系列基准测试。

According to a blog post published by Moonshot AI, the Kimi K3 model was specifically designed to handle long-horizon coding, knowledge work, and complex reasoning tasks. To evaluate its capabilities in these specialized domains, the startup conducted a series of benchmark tests comparing its performance against leading international competitors.

在月之暗面公布的自测基准测试结果中，Kimi K3在少数评估项目中展现出了超越美国顶尖系统的能力。具体而言，在Program Bench和SWE Marathon等测试中，Kimi K3的表现超越了Anthropic的Claude Fable 5以及OpenAI的GPT-5.6 Sol。这两款被超越的系统均是目前美国人工智能实验室所研发的最先进的AI系统，而Kimi K3的自测成绩表明其在特定专业任务上已具备极强的竞争力。

In the self-reported benchmark results released by the company, Kimi K3 demonstrated superior performance over two of the most advanced AI systems developed by prominent US laboratories. Specifically, the model outperformed Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol on a few select evaluations, including Program Bench and SWE Marathon. These self-reported victories highlight the model's technical focus and its competitive edge in executing highly specialized programming and reasoning tasks.

月之暗面推出Kimi K3这一全新模型，被广泛视为中国人工智能实验室正在缩小与美国顶尖平台技术差距的最新迹象。长期以来，以OpenAI和Anthropic为代表的美国科技巨头在生成式人工智能领域占据着主导地位。然而，随着中国本土AI先锋企业在技术上的快速迭代，这种单方面领先的格局正在发生改变。

The release of Kimi K3 by Moonshot AI serves as the latest sign that artificial intelligence laboratories in China are closing the technology gap with top-tier US platforms. For a long time, American tech giants such as OpenAI and Anthropic have maintained a dominant position in the global generative AI sector. However, the rapid progress of Chinese developers is actively challenging this established hierarchy.

根据相关行业分析，Kimi K3在特定基准测试中的表现，证明了中国研发的系统已经能够与美国最先进的AI平台并驾齐驱。尽管美国实验室在整体生态和部分前沿探索上仍具优势，但中国初创公司通过推出超大规模的开源模型，不仅降低了全球开发者的使用门槛，也展示出其在复杂推理和专业编程任务上追赶并挑战美国同行的强劲实力。

According to industry observations, the new model's ability to perform on par with some of the most advanced US systems in specific benchmarks highlights the accelerating capabilities of Chinese AI pioneers. By offering a highly complex, open-source alternative, Chinese startups are not only empowering the global developer community but also demonstrating their capacity to directly compete with leading American laboratories in specialized reasoning and coding tasks.

在世界人工智能大会上，总部位于杭州的强脑科技（BrainCo）展示了其研发的“脑控”机器人平台，展示了中国在具身智能领域的新进展。具身智能将人工智能与机器人等物理实体结合，使其能感知、推理并与物理世界交互。该平台允许用户仅凭思维，通过佩戴脑电图（EEG）头戴设备来控制机器人。

At the World AI Conference, Hangzhou-based technology firm BrainCo unveiled its brain-controlled robot platform, showcasing China's latest progress in the embodied AI sector. Embodied AI integrates artificial intelligence into physical machines, such as robots, enabling them to perceive, reason, and interact with the physical world. BrainCo's new platform allows users to control physical machines using only their thoughts through electroencephalogram (EEG) headsets.

据强脑科技介绍，该平台通过用户佩戴的脑电图头戴设备捕捉脑信号，并利用人工智能算法解码这些神经信号以识别用户意图，将其直接转化为机器人动作。该平台兼容人形机器人、机械臂和机器狗等多种第三方硬件。例如，仅凭大脑信号，机械臂就可以被指挥去抓取杯子和捡起苹果。

According to BrainCo, the platform operates by capturing brain signals via the EEG headset worn by the user. AI-based algorithms then decode these neural signals to identify the user's intentions, translating them directly into robotic actions. The platform is designed to be compatible with a wide range of third-party hardware, including humanoid robots, robotic arms, and robotic dogs. For instance, a robotic arm can be directed to grasp a cup and pick up an apple based solely on the user's brain signals.