[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-qwen-3-billion-downloads-open-weights":3,"news-related-aa4d3e55-383d-4855-9965-cc6a4d2e38a7":38},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"aa4d3e55-383d-4855-9965-cc6a4d2e38a7","Qwen 下载量 6 个月破 30 亿:开源模型的「默认底座」第一次换成了中国厂商","8 月 15 日阿里巴巴宣布,Qwen 系列开放权重模型过去六个月全球累计下载量突破 30 亿,按 Hugging Face 指标超过 Meta Llama(约 2.27 亿)与 Google Gemma(约 4.18 亿),成为全球下载量最大的开源模型家族。生态侧已有 460 多个开源模型、30 万个衍生模型。与此同时,Reuters 报道阿里正推动大型商用用户收入分成计划,「开源」的商业边界正在被重新定义。","# Qwen 下载量 6 个月破 30 亿:开源模型的「默认底座」第一次换成了中国厂商\n\n8 月 15 日,阿里巴巴宣布:Qwen 系列开放权重模型在过去六个月的全球累计下载量突破 **30 亿次**。据多方转引的 Hugging Face 指标,这个数字把 Meta 的 Llama 家族(约 2.27 亿次)和 Google 的 Gemma 家族(约 4.18 亿次)都甩在了身后——Qwen 成为全球下载量最大的开放权重模型生态([Bloomberg 报道](https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-08-15\u002Falibaba-ai-models-hit-3-billion-downloads-passing-meta-google))。\n\n## 不只是数字:生态的量级差\n\n支撑这个里程碑的是一套完整的模型矩阵:Qwen 家族目前已开源 **460 多个模型**,社区基于它们创建的衍生模型超过 **30 万个**。这不是某一款旗舰模型爆红的结果,而是把全尺寸段(从边缘端可跑的小模型到 2.4 万亿参数的 Qwen3.8-Max)都铺开之后的生态效应。\n\nHugging Face 的一份报告把 Qwen 称为「开放 AI 生态最大的地基之一」,并指出它「已经成为开发者决定微调和部署什么模型时的默认工作流的一部分」。这句话的分量比下载量本身更重——「默认底座」意味着后来者做微调、做 Agent、做行业应用时,第一反应是拿 Qwen 起步。\n\n## 为什么是 Qwen,不是 Llama\n\nNewsBytes 的分析点出了几个结构性原因:其一,中国厂商持续推出能力强、成本低、易于改造的模型,Qwen 是集大成者;其二,阿里把 Qwen 铺到了自家云平台的企业客户那里,覆盖东南亚和非洲市场,触达面比多数竞争对手更宽。与此同时,阿里也在跑赢国内同行——DeepSeek、Moonshot AI、MiniMax 的下载量均落后一档。作为回应,Meta 和 NVIDIA 等美国厂商正在加速推出自己的开放模型。\n\n在开源权重阵营里,「下载量 + 衍生模型数」正在成为衡量影响力的通用货币。这个指标第一次指向中国公司,而且是断层式领先——30 亿对 2.27 亿,超过一个数量级。\n\n## 另一面:收入分成计划,「开源」的定义正在被改写\n\n但这个节点上还有另一条线索值得放在一起看。Reuters 8 月 7 日报道,阿里计划要求 Qwen3.8-Max 权重的大型商用用户,从基于该模型的产品或服务收入中拿出一部分与阿里分成;针对的是大型企业用户,研究者和小开发者不受影响,具体比例当时仍在谈判中。这条路并非阿里首创——Moonshot AI 的 Kimi K3 许可证据报对年收入超过约 2000 万美元的公司寻求最高 30% 的收入分成。\n\n也就是说:一边是史上最大的开放权重生态,另一边是「大商用要分成」的新规则。「免费开源」的经典语义,正在被改写成「小规模免费、大规模付费」的分层结构。这是开源 AI 经济学的一次结构性变化,不再只是许可证条款的细节。\n\n## 所以呢\n\n30 亿下载确认了一件事:全球开发者的默认模型底座,第一次由一家中国公司提供。但接下来更值得盯的不是下载量,而是那个分成计划最终落在什么比例、什么门槛上——它将决定「开放权重」这趟车,对大公司到底是免费高速,还是收费国道。对开发者的建议很直接:小规模用,当下红利真实存在;若要押上生产系统,先把许可证条文逐字读完。\n\n参考:[Bloomberg 原报道](https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-08-15\u002Falibaba-ai-models-hit-3-billion-downloads-passing-meta-google) · [NewsBytes 分析](https:\u002F\u002Fwww.newsbytesapp.com\u002Fnews\u002Fscience\u002Falibaba-s-qwen-models-outpace-google-meta-in-downloads\u002Fstory)","https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-08-15\u002Falibaba-ai-models-hit-3-billion-downloads-passing-meta-google","e788d5af-1efa-40df-9646-6a9d702af265",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":22,"name":23,"slug":23,"description":14,"color":14},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"6f3722d1-f277-4ea1-b962-584a0d4507c3","en","Qwen tops three billion downloads, the open default flips","On August 15, Alibaba announced that its Qwen family of open-weight models has surpassed 3 billion cumulative global downloads in the past six months. Per Hugging Face metrics cited by Bloomberg, this eclipses Meta Llama (~227M) and Google Gemma (~418M), making Qwen the most-downloaded open-weight model family in the world. The ecosystem now spans 460+ open-sourced models and over 300,000 derivatives. Meanwhile, Reuters reports Alibaba is pushing a revenue-sharing plan for large commercial users, redefining the commercial boundary of open weights.","# Qwen Passes 3 Billion Downloads in Six Months: The Default Open-Weight Foundation Is Now Chinese\n\nOn August 15, Alibaba announced that its Qwen family of open-weight models has passed **3 billion cumulative global downloads** over the past six months. According to Hugging Face metrics cited across outlets, that figure leaves Meta's Llama family (~227 million) and Google's Gemma family (~418 million) far behind — Qwen is now the most-downloaded open-weight model ecosystem in the world ([Bloomberg report](https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-08-15\u002Falibaba-ai-models-hit-3-billion-downloads-passing-meta-google)).\n\n## Not Just a Number: The Ecosystem Scale Gap\n\nBehind the milestone is a full-spectrum model matrix: the Qwen family has open-sourced **more than 460 models**, and the community has created **over 300,000 derivative models** on top of them. This is not the result of one flagship model going viral — it is the ecosystem effect of covering every size tier, from edge-runnable small models to the 2.4-trillion-parameter Qwen3.8-Max.\n\nA Hugging Face report called Qwen \"one of the largest foundations of the open AI ecosystem\" and noted that it \"has become part of the default workflow for developers deciding what models to fine-tune and deploy.\" That sentence carries more weight than the download count itself — being the default foundation means that when builders fine-tune, build agents, or ship industry applications, their first instinct is to start from Qwen.\n\n## Why Qwen, Not Llama\n\nNewsBytes' analysis points to several structural reasons. First, Chinese vendors keep shipping models that are capable, cheap, and easy to adapt, with Qwen as the archetype. Second, Alibaba extended Qwen to enterprise customers on its own cloud platform across Southeast Asia and Africa, giving it wider reach than most competitors. At the same time, Alibaba is outpacing domestic peers — DeepSeek, Moonshot AI, and MiniMax all trail in downloads. In response, US giants like Meta and NVIDIA are racing to launch their own open models.\n\nWithin the open-weight camp, \"downloads + derivative count\" has become the standard currency of influence. For the first time, that metric points to a Chinese company — and by a landslide: 3 billion versus 227 million is more than an order of magnitude.\n\n## The Other Side: Revenue Sharing Is Rewriting What \"Open\" Means\n\nBut there is another thread worth placing alongside this milestone. Reuters reported on August 7 that Alibaba plans to require large commercial users of the Qwen3.8-Max weights to share a portion of the revenue generated from products or services built on the model. The plan targets major enterprise users; researchers and small-scale developers are unaffected, and the exact percentage was still under negotiation at the time of reporting. Alibaba is not first on this path — Moonshot AI's Kimi K3 license reportedly seeks a revenue share of up to 30% from companies with annual revenue above roughly $20 million.\n\nIn other words: on one side, the largest open-weight ecosystem in history; on the other, a new rule of \"large commercial use pays.\" The classic semantics of \"free and open source\" are being rewritten into a tiered structure of \"free at small scale, paid at large scale.\" This is a structural change in the economics of open AI — no longer a footnote in license terms.\n\n## So What\n\nThe 3-billion-download figure confirms one thing: for the first time, the default model foundation for developers worldwide is provided by a Chinese company. But what to watch next is not the download count — it is where the revenue-sharing plan finally lands: what percentage, what threshold. That will determine whether the \"open weights\" express is a free highway or a toll road for large companies. The advice for developers is straightforward: at small scale, the upside is real today; before betting a production system on it, read every line of the license first.\n\nReferences: [Bloomberg report](https:\u002F\u002Fwww.bloomberg.com\u002Fnews\u002Farticles\u002F2026-08-15\u002Falibaba-ai-models-hit-3-billion-downloads-passing-meta-google) · [NewsBytes analysis](https:\u002F\u002Fwww.newsbytesapp.com\u002Fnews\u002Fscience\u002Falibaba-s-qwen-models-outpace-google-meta-in-downloads\u002Fstory)","qwen-3-billion-downloads-open-weights","2026-08-15T23:20:00Z","2026-08-15T23:10:13.277735Z","2026-08-15T23:10:13.277743Z",true,"agent",73,{"items":39},[40,45,50,55,60,65],{"id":41,"title":42,"news_slug":43,"published_at":44},"0d8fdf45-4585-47c0-9e78-3652e318b156","Apple Intelligence 中国版落地:通义千问接管语言 AI,百度负责视觉搜索","apple-intelligence-china-qwen-baidu-2026","2026-08-25T12:00:00+00:00",{"id":46,"title":47,"news_slug":48,"published_at":49},"1844afb1-3a1c-4acd-9e4c-f5e2792a2018","下载免费不等于商用免费：HF Summer 2026 隐藏的开源前沿许可证分水岭","frontier-license-shift-hf-summer-2026","2026-08-23T12:30:00+00:00",{"id":51,"title":52,"news_slug":53,"published_at":54},"4bb31ede-b9c4-4762-86ae-9d3b008557ca","Hugging Face Summer 2026 报告:Qwen 拿下 15 万衍生模型, GGUF 仓库一年涨 464%","hugging-face-state-of-open-models-summer-2026","2026-08-18T02:00:00+00:00",{"id":56,"title":57,"news_slug":58,"published_at":59},"b863d01a-dfdf-41c9-8266-e4602e58bde3","Qwen3.8-Max 开源权重落地:砍掉视觉与 1M 上下文,许可证换成收入分成","qwen3-8-max-open-weights-stripped-relicense","2026-08-23T13:30:00+00:00",{"id":61,"title":62,"news_slug":63,"published_at":64},"9389d1ed-dd2d-41cb-bbc5-9a543e2b2f71","开源报告里的「参数天花板」分水岭:中国实验室把上限拉到2.78T,美国还在130B徘徊","hf-summer-2026-china-open-weight-parameter-ceiling","2026-08-20T06:00:00+00:00",{"id":66,"title":67,"news_slug":68,"published_at":69},"b2c169c6-5150-4423-8073-bf480a2d8745","腾讯 UniPert-G2CP 登《Cell》主刊：把基因扰动和化学扰动塞进同一个语义空间","tencent-unipert-g2cp-cell-virtual-cell","2026-07-31T07:49:00+00:00"]