模型
别名：通义千问
阿里云大语言模型系列，Holo3.1 基于其架构设计。
已跟踪 30 条高相关材料
最近变化
2026-06-02 · Holo3.1 在 AndroidWorld 上 35B-A3B 模型准确率从 67% 提升至 79.3%
为什么值得关注
Qwen 被反复提及时，通常意味着它正在影响产品路线、开发者工作流或 AI 产业判断。这个页面把分散材料合并成一个可持续更新的观察入口。
DeepSeek 的 10 万亿美元大战略
宝玉的分享 · 9.2 分
DeepSeek通过多项技术创新大幅降低大模型推理中的KV缓存需求，推动中国AI硬件生态发展，目标打造价值10万亿美元的产业巨兽。
CyberSecQwen-4B: Why Defensive Cyber Needs Small, Specialized, Locally-Runnable Models
Hugging Face Blog · 8.8 分
小型专用模型如 CyberSecQwen-4B 在防御性网络安全任务中表现优于大型通用模型，且支持安全本地部署，无需 API 成本。
#558.AI时代的个人革命：Garry Tan 谈开源 AI、创业信仰、创伤动力
跨国串门儿计划 · 8.7 分
Garry Tan 认为AI正开启下一次个人计算革命，开源Agent与个人AI将赋予普通人前所未有的创造能力；YC核心信条是‘做出人们想要的东西’；创业者需以真诚感知力与主观能动性将创伤转化为创造力。
已收录 30 条与 Qwen 相关的内容，按评分排序。
DeepSeek reduces KV cache requirements through innovations, driving China's AI hardware ecosystem toward a $10 trillion industry.
入选理由：DeepSeek V4 Pro仅需5.48GB HBM，相比GLM5的60GB和Qwen3-235B-A22B的89GB显著节省显存
Small specialized models like CyberSecQwen-4B outperform larger generalists in defensive cyber tasks while enabling secure local deployment without API costs.
入选理由：CyberSecQwen-4B 在 CTI-MCQ 测试中以 0.5868 准确率超越 Cisco 8B 模型 8.7 个百分点。
Garry Tan argues that AI is triggering the next personal computing revolution, where open-source Agents and personal AI will empower ordinary people with unprecedented creative capacity; YC’s core tenet is “make something people want”; entrepreneurs must convert trauma into creativity through authentic perception and strong agency.
入选理由：Garry Tan 提出‘个人AI必须由自己拥有和控制’，并正在开发 G Brain——整合邮件、日历、联系人与笔记的个人知识记忆系统。
Holo3.1 is Hugging Face's new computer-use agent model supporting cross-platform, multi-framework deployment and first releasing quantized weights (FP8/Q4 GGUF/NVFP4) for local inference.
入选理由：Holo3.1 在 AndroidWorld 上 35B-A3B 模型准确率从 67% 提升至 79.3%
Databricks 提供了一个可靠的 LLM 推理平台，支持大规模多租户系统，通过先进的硬件和软件优化实现高可用性和低延迟。
入选理由：Databricks 平台支持多种前沿模型，包括开源和专有模型。
本文介绍了如何通过显式缓存优化Qwen模型的使用，包括缓存的工作原理、实现方法和最佳实践，帮助用户提高效率并降低成本。
入选理由：显式缓存可以显著减少重复请求的处理时间，提高响应速度。
Qwen在自主执行过程中，通过连续运行约35小时，进行了1158次工具调用，完成了432次内核评估，自主编写、编译、分析和迭代改进了Extend Attention Kernel，实现了10.0倍的几何提升。
入选理由：Qwen在35小时内自主执行，进行了1158次工具调用和432次内核评估。
Qwen on X highlights the latest developments in AI, particularly the Qwen3.7 Plus Preview and its impact on the Vision Arena.
入选理由：Qwen's Qwen3.7 Plus Preview enhances performance with new features, making it competitive in the lab rankings.
GB 200s improve the prefill and decode disaggregation efficiency for large MoE models like Qwen, significantly enhancing throughput compared to the Hopper platform.
入选理由：GB 200s 在高吞吐量推理方面比 Hopper 更适合大型 MoE 模型。
Qwen 3.7 Max ranks fourth in Arena Coding Agent, surpassing GLM-5.1 and matching Claude Opus 4.6 in agentic web development tasks.
入选理由：Qwen 3.7 Max 排名第四
Qwen3.7-Max has reached position #4 on Code Arena, tying with Claude Opus 4.6, making it the top-ranked Chinese lab!
入选理由：Qwen3.7-Max 在 Code Arena 上排名第 4，超越 GLM-5.1。
Using FireworksAI Agent to automate LLM fine-tuning demonstrates the feasibility of self-improving AI systems, enabling model iteration through natural language interaction. Future recursive self-improvement systems could revolutionize knowledge discovery and end-to-end research automation.
入选理由：FireworksAI Agent已实现LLM微调自动化，成功优化Qwen模型输出风格以适配PaperWiki知识库
林俊旸离职阿里后创立的新公司种子轮融资目标估值高达20亿美元，其核心理念是推动AI从推理式思考转向智能体式思考，即让模型为了行动而思考，强调环境设计、轨迹采样和编排工程的重要性。
入选理由：林俊旸离职阿里后创立的新公司种子轮融资目标估值高达20亿美元。
Claude model dominates the Code Arena frontend leaderboard, with OpenAI and Gemini rankings declining.
入选理由：Claude模型占据了前五名，Opus 4.7 Thinking以30分优势领先。
Qwen3.5 达到 580 tps 的记录性突破，得益于 TokenSpeed 引擎和合作伙伴的优化。
入选理由：Qwen3.5 在 TokenSpeed 引擎上实现 580 tps 的性能。
At the AMD AI Developer Conference in Shanghai, CEO Lisa Su stated that AI competition is shifting from model capabilities to systems engineering and full-stack optimization. Developers need a deployable, optimizable, and continuously evolving engineering system. AMD, centered on its ROCm open-source platform, provides full-stack computing power from cloud to edge, while continuously strengthening its developer ecosystem in China.
入选理由：AI行业竞争焦点正从模型能力转向系统工程与全栈优化能力，开发者需要可落地、可优化、可持续演进的工程体系
MiniCPM-V 4.6 is a compact 1.3B parameter multimodal vision-language model using SIGLIP visual encoder and Qwen language model architecture, supporting image, document and video inputs for edge device deployment.
入选理由：模型仅 13 亿参数，支持 262K 上下文窗口处理多图像和视频
Hermes Agent is a local AI assistant framework that combines OpenRouter and Qwen models for efficient reasoning, supports memory layers, Obsidian integration, and local automation scripts, running on Android devices.
入选理由：Hermes Agent 使用 Qwen 3.6+ 模型实现推理成本降低至原价的十分之一
Hermes Agent is a local-running intelligent agent system that combines OpenRouter and Qwen models for efficient reasoning, supports memory layers and automation scripts, and can run on Android devices.
入选理由：Hermes Agent 使用 Qwen 3.6+ 实现低成本高效率推理，将 token 成本从 $100 降至 $10。
Alibaba Cloud's Qwen series multi-size models are now available on the SiliconFlow platform.
入选理由：Qwen3.5-397B-A17B是最大参数模型
Qwen3.7 preview version is now on Arena, Alibaba's vision ranking rises to fifth, and the model series will be released soon.
入选理由：Qwen3.7-Plus-Preview在Arena视觉竞技场排名第五，整体排名第十六
Geekbb announced that Alibaba's Qwen3.6-Plus has gone live on Nous Portal and is free for a limited time. This marks a new phase in the collaboration between Qwen and Nous Research.
入选理由：阿里巴巴的 Qwen3.6-Plus 已经上线 Nous Portal 并且限时免费。
Qwen released a demo video of Browser Agent on X, but no technical details are provided, only showing examples of automated browser operations.
入选理由：Qwen于2026年6月1日在X平台发布Browser Agent演示视频。
Qwen 3.7 Max is now supported in Hermes Agent.
入选理由：Qwen 3.7 Max 在 Hermes Agent 中支持
Qwen3.7-Plus is a multimodal agent model from Qwen that unifies vision and language capabilities, supporting GUI/CLI operations and coding tasks, now available via Alibaba Cloud API.
入选理由：Qwen3.7-Plus支持多模态交互，统一GUI和CLI操作处理视觉和文本任务。
Qwen3.6-Plus is now free on Nous Portal for a limited time, allowing users to access over 300 models with a single subscription, enjoy exclusive discounts, and simplified billing.
入选理由：Qwen3.6-Plus 在 Nous Portal 上限时免费。
The article is an announcement from Alibaba's Qwen team recruiting ambassadors, offering early model access and API credits.
入选理由：招募开发者和社区领袖担任大使
Qwen announced Demo2: Multimodal Interactive Hybrid Agent, but no technical details or implementation principles were provided.
入选理由：Qwen于2026年6月1日在X上宣布了Demo2: Multimodal Interactive Hybrid Agent。
Orange AI on X shared a tweet about Anthropic distilling Chinese models, indicating evidence that Claude distilled Kimi and Qwen, but the responsibility is questionable.
入选理由：Anthropic 蒸馏了中国的 Kimi 和 Qwen 模型。
This post spreads a clearly false rumor: Claude is developed by Anthropic, Kimi by Moonshot AI; there is zero technical linkage—this is a typical AI misinformation case.
入选理由：Claude由Anthropic公司研发，与月之暗面的Kimi模型无任何训练或架构关系
