最新资讯
dev.to 我们在 DolphinBench 上测试记忆检索:纯关键词搜索暴露了自身回归 dev.to 别再把 Agent 写成 while 循环,把它建模为状态机 bloomberg.com Rightmove 与 Scout24 借 AI 产品应对低迷房市并上调展望 stackoverflow.blog 欢迎来到芯片时代:半导体制造、AI 数据中心与芯片设计的未来 www.youtube.com AI Agent 总犯同样的错:CentralAgent AI 将记忆重构为经验系统 pub.towardsai.net Netflix 科学家建议重采样 LLM 评判:5 次仅将误差棒缩小 13% pub.towardsai.net Jev 作为 LLM 前的静默层:构建提示感知的模型路由器 reddit.com OpenAI 称 Astra 仅满足 10% 安全任务标准 dev.to FreeToken:在自有 GPU 上运行前沿 MoE 模型及其内存需求 dev.to 当 AI Agent 说“完成”时,究竟完成了什么? dev.to 部署 Underdog Bark 3.0 Pro:逐字转写、自定义后端与分块设置 dev.to 为艰难对话构建 AI:五个 Agent、一条安全正则与一位盲评监督者
多元动态
登录 注册
智能体 www.aibase.com ·

月之暗面 Kimi K3.1 前端标识泄露:百万上下文与三档推理强度

原文 我的收藏

开发者发现月之暗面 API 平台后端模型注册表出现 kimi-k3-1 标识符,且接口探测可调用;当晚 Kimi 开放平台也预告 K3.1(或代号 K311111)支持最高 100 万 token 上下文。泄露配置显示三档推理强度 Low、High、Max,并可能引入原生 Agent 模式与 Swarm 多智能体协作及搜索、批处理任务模式。API 定价暂与 K3 相同,可能为占位。模型尚未正式发布,相关能力与定价仍待官方确认。

AI推荐理由 内容基于开发者发现 API 后端出现 kimi-k3-1 标识符及官方开放平台预告,属可验证的泄露线索,但模型未正式发布,价格、能力均为推测,来源为 AIbase 转述而非官方公告,事实明确度有限。兴趣点在于百万 token 上下文、三档推理强度和 Swarm 多智能体协作若属实,对 Agent 与长上下文应用有参考价值,但需保留不确定性。

62

正文

The strongest model of Moon's Dark Side, Kimi K3.1, has leaked frontend identifier; one million tokens context with three levels of reasoning strength is imminent AIbase基地 Published inAI News · 3 min read · Sep 29, 2026 11 The next flagship from Moonshot AI was unearthed by developers themselves. On September 28, multiple developers noticed the identifier "kimi-k3-1" appearing in the backend model registry of Moonshot AI's API platform. This identifier not only appeared in the list but also passed the interface probe and could indeed be called. Later that evening, the official Kimi open platform also posted a preview of the K3.1 model, or codenamed "K311111," stating support for a context window of up to 1 million tokens. From the leaked configuration, Kimi K3.1 is expected to reach a maximum context window of 1 million tokens and offers three adjustable levels of reasoning intensity: Low, High, and Max. More notably, it may introduce native Agent intelligence mode and Swarm multi-agent collaboration, along with task modes such as search and batch processing. Developers have analyzed that splitting the reasoning intensity into multiple levels might correspond to different computing resource allocations and billing models—users can choose Low for cost savings or Max for optimal quality, upgrading from "a single model working alone" to "a model commanding a group of models." Price details remain undisclosed for now. The Kimi open platform shows that the API pricing of K3.1 is the same as the current K3, although this might just be a placeholder since the model has not been officially announced. Regardless of the final pricing, from the detection of the backend identifier being adjustable to the official platform's proactive announcement, K3.1 has already crossed the threshold of recent releases—when Moonshot AI includes a million-token context, three levels of reasoning intensity, and multi-agent collaboration in its configuration, the focus of the next round of large model competition is shifting from "who is better at chatting" to "who can handle tasks on their own." KimiK3.11MtokensAgent intelligent body modeSwarm multi-agent collaboration This article is from AIbase Daily Scan to view Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications. —— Created by the AIbase Daily Team © Copyright AIbase Base 2024, Click to View Source -

来源报道 · 1 篇 · 以下链接均前往原始媒体

← 返回精选动态 原文

更多内容