Daily Tech Briefing
AI 科技速览
每天 5 分钟内学习 AI。获取最新的人工智能新闻,理解其重要性,并学习如何将其应用于您的工作。
VentureBeat ML · 2026/7/28 02:08:51

Kimi K3's full weights are here, but they're 'open' with a caveat: What enterprises should know
AI 中文解读
核心亮点:中国AI公司月之暗面今天开放了最强模型Kimi K3的完整权重,但企业要小心——它不是传统意义上的“开源”,而是带有特殊使用条款的“开放”。
通俗解读:Kimi K3是一个超级大的AI模型,参数多到2.8万亿,能一次处理100万字的文档,还能看懂图片和视频。现在任何人都可以下载它的完整“配方”(包括权重、代码和运行工具),自己部署或修改。不过,它附带的许可协议虽然允许免费用于商业,却额外要求大公司和AI服务商必须保留版权声明、遵守法律等,这比常见的Apache 2.0开源许可更严格。比如,你不能用它做违法的事,也不能随便甩锅给别人。
实际影响:对普通人来说,这意味着未来会有更多基于Kimi K3的本地AI应用出现——比如企业用它搭建自己的客服机器人或文档分析工具,不用每次都通过付费API调用,成本更低。但对大公司而言,用之前必须仔细审查协议条款,避免踩坑。整体上,这项发布降低了高端AI的自建门槛,但也提醒大家“免费”不等于“无限制”,商业使用前最好找法务把关。
Chinese AI startup Moonshot AI, creators of the increasingly famous "Kimi K" family of powerful, open AI models, today released the full weights for its largest and most performant version yet — Kimi K3. But enterprises evaluating the model should read the attached custom Kimi K3 usage license as carefully as the benchmark charts.VentureBeat previously covered Kimi K3 when it debuted through Moonshot's hosted API earlier this month, including its 2.8 trillion-parameter architecture, one million-token context window and frontier benchmark performance. Today's announcement completes that rollout with the release of the full model weights, a 47-page technical report documenting its training innovations and obstacles, and much of the infrastructure required to run the model independently.That includes the complete 2.8 trillion-parameter Mixture-of-Experts model, inference infrastructure, optimized attention kernels, MoE communication libraries and deployment components aimed at researchers and enterprise developers who want to self-host the system rather than consume it exclusively through an API. Moonshot is also releasing implementation support for ecosystems including vLLM and SGLang, alongside a technical report detailing the architectural innovations behind the model, including Kimi Delta Attention, Attention Residuals and Stable LatentMoE. Together, those techniques underpin what the company describes as the world's first open 3T-class model, activating 104 billion parameters from a pool of 896 experts while supporting native multimodal reasoning and a one million-token context window.While Kimi K3's license grants broad rights to developers and even enterprises to download, modify and deploy the model for commercial purposes — a win for those seeking low-cost, frontier-level AI they can control and run offline — it also imposes some added obligations on larger companies and AI service providers not found in traditional open-source licenses like Apache 2.0 or MIT.The key restrictions enterprises should understandHere's the text of the new Kimi K3 License in full:Permission is hereby granted, free of charge, to any person (the "Licensee") obtaining a copy of this software — including the model weights, parameters, configuration files, inference and training code, and associated documentation (collectively, the "Software") — to deal in the Software without restriction. This includes, without limitation, the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software; to run, deploy, fine-tune, or otherwise modify the Software and create derivative works from it; and to permit persons to whom the Software is furnished to do so, in each case subject to the following conditions: 1. The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Licensee's use of the Software must comply with applicable laws and regulations. 2. "Model as a Service" means giving a third party access to language model inference or fine-tuning (e.g., via API) in a manner that allows such third party to exercise meaningful control over the inputs, parameters, or training data. This does not include (a) end-user products with model capabilities solely embedded within specific features or harnesses, or (b) mere relaying of requests to models hosted by others. If the Licensee or any of its affiliates operates a Model as a Service business, and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months, the Licensee must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose. 3. If the Software (or any derivative works thereof) is used for any of the Licensee's commercial products or services that have more than 100 million monthly active
分享
阅读原文 ↗