AI News AI资讯 1d ago Updated 18h ago 更新于 18小时前 49

Alibaba's Qwen takes on Kimi K3 with open-weight Qwen 3.8, says model is "second only to Fable 5" 阿里巴巴Qwen以开源Qwen 3.8迎战Kimi K3,称模型“仅次于Fable 5”

Alibaba released Qwen 3.8, an open-weight multimodal model with 2.4 trillion parameters capable of processing images, videos, and documents. The Qwen team claims the model matches frontier capabilities and ranks second only to Fable 5, with significant improvements over Qwen 3.7-Max in coding and productivity tasks. Early access is provided via Alibaba’s Token Plan, Qoder, and QoderWork at 10% of the standard price, while full open weights are scheduled for imminent release. This move strategica 阿里巴巴发布开源权重模型 Qwen 3.8,参数量高达 2.4 万亿,定位为多模态大模型。 官方宣称该模型性能仅次于 Fable 5,并显著优于前代 Qwen 3.7-Max,尤其在代码和复杂生产力任务上。 这是 Qwen 团队首个参数量超过万亿的多模态模型,支持图像、视频和文档处理,但暂无公开基准测试结果。 此次发布被视为对 Moonshot AI(Kimi K3)商业策略的直接竞争,旨在通过开源优势抢占开发者生态。

75
Hot 热度
65
Quality 质量
70
Impact 影响力

Analysis 深度分析

TL;DR

  • Alibaba released Qwen 3.8, an open-weight multimodal model with 2.4 trillion parameters capable of processing images, videos, and documents.
  • The Qwen team claims the model matches frontier capabilities and ranks second only to Fable 5, with significant improvements over Qwen 3.7-Max in coding and productivity tasks.
  • Early access is provided via Alibaba’s Token Plan, Qoder, and QoderWork at 10% of the standard price, while full open weights are scheduled for imminent release.
  • This move strategically targets Moonshot AI’s Kimi K3, aiming to disrupt its customer acquisition funnel ahead of Moonshot’s potential IPO.

Why It Matters

This release marks a critical escalation in the open-weight LLM arms race, demonstrating that massive parameter counts (2.4T) can compete with proprietary frontier models. For developers and enterprises, it provides a high-capability, multimodal alternative to closed-source systems, potentially lowering barriers to entry for complex tasks like full-stack development and data analysis.

Technical Details

  • Model Architecture: Qwen 3.8 is a 2.4 trillion parameter model, identified as the team's first multimodal architecture exceeding 1 trillion parameters.
  • Multimodal Capabilities: Unlike previous text-only iterations, this model natively processes images, videos, and documents, expanding its utility in complex productivity workflows.
  • Performance Claims: The team asserts superior performance in coding and full-stack development compared to Qwen 3.7-Max, positioning it as a direct competitor to top-tier proprietary models.
  • Access Mechanism: Preliminary evaluation is available through Alibaba’s ecosystem (Token Plan, Qoder, QoderWork) at a discounted rate, with full open-weight distribution pending.

Industry Insight

  • Strategic Disruption: Alibaba is directly challenging Moonshot AI’s growth trajectory by offering a comparable open alternative before Moonshot’s anticipated IPO, potentially stifling its ability to convert free users into paying API customers.
  • Open vs. Closed Dynamics: The release reinforces the trend where open-weight models are becoming viable substitutes for proprietary APIs in enterprise settings, forcing closed-model providers to justify their pricing through unique features or reliability rather than just performance.
  • Multimodal Standardization: The shift toward trillion-parameter multimodal models suggests that future benchmarks will increasingly require integrated vision-language-text capabilities, pushing competitors to accelerate their own multimodal releases.

TL;DR

  • 阿里巴巴发布开源权重模型 Qwen 3.8,参数量高达 2.4 万亿,定位为多模态大模型。
  • 官方宣称该模型性能仅次于 Fable 5,并显著优于前代 Qwen 3.7-Max,尤其在代码和复杂生产力任务上。
  • 这是 Qwen 团队首个参数量超过万亿的多模态模型,支持图像、视频和文档处理,但暂无公开基准测试结果。
  • 此次发布被视为对 Moonshot AI(Kimi K3)商业策略的直接竞争,旨在通过开源优势抢占开发者生态。

为什么值得看

对于 AI 从业者和行业观察者而言,Qwen 3.8 的发布标志着超大规模开源模型在性能上已能逼近甚至挑战顶级闭源模型,加剧了开源与闭源阵营的技术博弈。同时,这一举动揭示了头部科技公司在 AI 商业化路径上的战略分歧,特别是针对 Moonshot AI 等新兴独角兽的上市计划构成了潜在的市场干扰。

技术解析

  • 模型规格:Qwen 3.8 拥有 2.4 万亿参数,是 Qwen 团队首款万亿级多模态模型,具备处理图像、视频和文档的能力。
  • 性能定位:官方声称其性能匹配前沿模型,仅落后于 Fable 5,并在全栈开发、数据分析及办公工作流等复杂任务中超越 Qwen 3.7-Max。
  • 访问方式:目前通过 Alibaba's Token Plan、Qoder 和 QoderWork 提供预览,价格为标准价格的 10%,完整开源权重预计“即将”推出。
  • 数据缺失:截至目前,官方尚未公布具体的基准测试(Benchmark)结果,性能声明主要基于内部评估。

行业启示

  • 开源对抗闭源的商业模式:阿里巴巴通过提供高性能开源模型,试图打破 Moonshot AI 等公司依靠 API 和封闭应用转化用户的商业闭环,开源成为争夺开发者注意力和生态主导权的关键武器。
  • 超大规模模型的竞争门槛:2.4 万亿参数的多模态模型表明,AI 行业的军备竞赛已进入“巨无霸”阶段,只有具备强大算力储备和资金实力的巨头才能维持在这一层级的竞争力。
  • 资本市场的影响:在 Moonshot AI 计划短期内 IPO 之际,Qwen 3.8 的发布可能影响投资者对独立 AI 创业公司估值逻辑的判断,市场将更加关注开源替代方案对专有模型商业价值的侵蚀风险。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Open Source 开源 Product Launch 产品发布 Research 科学研究