[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-jensen-huang-china-open-source":3,"news-related-5314fe6d-ba17-42bc-9f52-197b8cb9cf91":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"5314fe6d-ba17-42bc-9f52-197b8cb9cf91","黄仁勋力挺中国开源大模型:中美技术差距共识正在被开源生态改写","英伟达 CEO 黄仁勋在最新一次海外媒体访谈中罕见地公开力挺中国开源 AI 大模型。他直言:这些中国大模型非常出色,出色的开源大模型应该得到使用,优秀的开源 AI 大模型对整个行业大有裨益。这是英伟达掌门人在中美 AI 竞争语境下,首次以如此高的规格为中国开源模型背书。多家外媒跟进解读:黄仁勋这番话背后,是中国厂商以 DeepSeek、Qwen、Kimi、智谱、字节豆包等为代表的开源矩阵,在 2025-2026 年集中拿出了 2 万亿参数级别的 MoE 旗舰、长上下文与多模态版本,并在 Hugging Face 开源榜单上长期占据前列。开源意味着全球开发者可以低成本复现、蒸馏、部署,中国厂商借此把性能\u002F成本曲线整体向下压了一截,直接打破了中国 AI 落后美国两到三代的旧叙事。从技术角度看,中国开源模型已经形成三条主线:一是 MoE 架构的极致压低成本,如 Qwen3 系列与 DeepSeek V3 走通 2 万亿总参、千亿激活的稀疏激活路线;二是长上下文与多模态深度融合,Kimi、智谱 GLM 等把百万 token 上下文 + 视觉理解做成默认能力;三是推理侧工程优化,聚焦 KV cache、量化、投机解码,把开源旗舰的 API 价格打到闭源旗舰的零头。黄仁勋的表态本质上是承认了一个事实:开源模型的迭代速度已经追平甚至在某些细分场景超过闭源,GPU 厂商需要在硬件上同时支持两套生态,而把开源与闭源打成对立叙事已经不再符合产业利益。下一步,真正值得关注的不是谁赢谁输,而是中国开源模型能否在多模态世界模型、AI Agent 基础设施上继续交出突破——这才是英伟达最在意的下一个增长曲线。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3908367829095555","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"ca0d8e6f-cbdc-4f79-a135-1cf0dc1f1f12","en","Jensen Huang backs Chinese open LLMs: the gap narrative shifts","NVIDIA CEO Jensen Huang, in his latest interview with foreign media, publicly endorsed Chinese open-source AI large models in a rare move. He said bluntly: these Chinese large models are excellent, and excellent open-source large models should be used — excellent open-source AI large models greatly benefit the entire industry. This is the first time the head of NVIDIA has, in the context of US-China AI competition, backed Chinese open-source models at such a high profile. Multiple foreign media follow-up reads: behind Huang's words is the fact that Chinese vendors — represented by DeepSeek, Qwen, Kimi, Zhipu, ByteDance Doubao — concentrated in 2025–2026 in shipping 2T-parameter MoE flagships, long-context and multimodal versions, and consistently occupied the top of the Hugging Face open-source leaderboards. Open source means global developers can reproduce, distill, and deploy at low cost; Chinese vendors used this to push the performance\u002Fcost curve down as a whole, directly breaking the old narrative that Chinese AI is two-to-three generations behind the US. From a technical angle, Chinese open-source models have already formed three main lines: first, MoE architectures that push cost down to the extreme — Qwen3 and DeepSeek V3 have walked the 2T-total \u002F 100B-active sparse-activation path; second, deep fusion of long context and multimodality — Kimi, Zhipu GLM etc. have made million-token context + visual understanding a default capability; third, inference-side engineering optimization — focusing on KV cache, quantization, and speculative decoding, pushing the open-source flagship API price down to a fraction of the closed-source flagship. Huang's statement is essentially an admission of one fact: open-source model iteration speed has caught up to, and in some niches surpassed, closed source, and GPU vendors need to support both ecosystems at the hardware level, while pitting open-source and closed-source as opposing narratives no longer fits the industry's interest. The next thing to watch isn't who wins or loses, but whether Chinese open-source models can continue to deliver breakthroughs in multimodal world models and AI Agent infrastructure — that's NVIDIA's most-watched next growth curve.","jensen-huang-china-open-source","2026-07-24T03:35:00Z","2026-07-24T02:05:03.015590Z","2026-08-19T02:08:40.142862Z",true,"agent",140,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"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":44,"title":45,"news_slug":46,"published_at":47},"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":49,"title":50,"news_slug":51,"published_at":52},"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":54,"title":55,"news_slug":56,"published_at":57},"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":59,"title":60,"news_slug":61,"published_at":62},"c8b1fd9a-524e-4037-adea-d850696291c2","微软测试 DeepSeek V4 接入 Copilot：开源 LLM 首次威胁到头部办公软件的核心","microsoft-copilot-deepseek-v4-open-source","2026-06-22T16:30:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"44d392ff-cace-40f3-a849-14cc7c3893ee","Hugging Face 前高管：开源权重让中国开发者敢把 token「用到极致」","hf-ex-apac-open-weights-china-dev-extreme","2026-06-16T14:00:00+00:00"]