前沿模型交锋
Kimi K3震动前沿. Moonshot AI的开源权重Kimi K3与顶级闭源模型竞争,在部分基准测试上超越Claude Fable 5,重新点燃了开源与闭源之争。
- Kimi's open model K3 nears GPT-5.6 Sol and Fable 5 while signaling the end of super cheap Chinese AI(在新标签页中打开)
- Kimi K3, and what we can still learn from the pelican benchmark(在新标签页中打开)
- [AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing(在新标签页中打开)
- Kimi K3 weights to be released on the 27th.(在新标签页中打开)
- Kimi K3 ranks #1 on @AfterQuery's SpreadsheetBench 2, surpassing Claude Fable 5(在新标签页中打开)
- Kimi: Threat or menace?(在新标签页中打开)
- kimi.ai teasing a video with lots of 3's in it(在新标签页中打开)
- [AINews] not much happened today(在新标签页中打开)
- Kimi moment. I think the writing is on the wall for Anthropic and OpenAi(在新标签页中打开)
- Kimi K3 is currently at the top of the leaderboard for Text Arena filtered for science queries.(在新标签页中打开)
- Kimi K3 (max) beats Sonnet 5 on Simple Bench(在新标签页中打开)
- Kimi K3 is top of nextjs eval(在新标签页中打开)
- Just like Deepseek, China's Kimi K3 is forcing Western AI labs to question their compute advantage(在新标签页中打开)
- Kimi K3 🌕, Gemini 3.5 delayed ⏳, crushing ARC-AGI 3 🤖(在新标签页中打开)
GPT-5.6:力量与风险. GPT-5.6系列带来了编程式工具调用和并行子代理,但它无意中删除文件和数据库的倾向凸显了沙盒化的必要性。
- The Sequence Radar #893: Last Week in AI: GPT-5.6, Grok 4.5, Muse Spark 1.1 and the Post-Chatbot Stack(在新标签页中打开)
- OpenAI’s new flagship model deletes files on its own, people keep warning(在新标签页中打开)
- GPT-5.6 is deleting user files when given full access, and OpenAI says it shouldn't but did(在新标签页中打开)
- [AINews] Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??(在新标签页中打开)
Claude Fable 5扩展. 在GPT-5.6和Kimi K3的压力下,Anthropic改变了计划,将Fable 5保留在付费方案中,直接竞争可及性。
- Fable gets another bump(在新标签页中打开)
- Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war(在新标签页中打开)
- Anthropic slashes Claude Fable 5 limits in Max and Team Premium and pushes Pro users toward API pricing(在新标签页中打开)
- Claude make Fable 5 permanent(在新标签页中打开)
DeepSeek V4浮现. DeepSeek正在预告V4,提供低成本API和开放权重,而社区优化已使其闪存变体在消费级GPU上以可用速度运行。
智能体进入真实工作流
浏览智能体到来. Anthropic为Claude Code内置了带安全分类器的网络浏览器,Cursor推出了通用智能体,推动编码助手走向自主任务执行。
智能体处理真实交易. DoorDash的智能体架构将转化率提高了24%,Stripe的基准测试显示智能体可以编写集成代码但常在验证环节失败,凸显了可靠性差距。
生产级智能体受挫. 大多数企业智能体仍然是聊天机器人;54%的公司发生过智能体安全事件,专家认为智能体需要类似微服务的云原生操作原语。
- How to Debug Coding Agents with LangSmith Traces(在新标签页中打开)
- The agent security gap: 54% of enterprises have already had an AI agent incident, and most still let agents share credentials(在新标签页中打开)
- The AI context gap: Enterprise AI organizations have a trust problem, not a retrieval problem — and most are still building the fix(在新标签页中打开)
- The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem — and most are shipping to production anyway(在新标签页中打开)
- Cloud Native Infrastructure Emerges as the Foundation for Trustworthy Agentic AI(在新标签页中打开)
- QCon AI Boston: Production AI Moves Beyond Prompts to Platforms, Harnesses, and Evals(在新标签页中打开)
AI基础设施、资金与数据中心竞赛
数据中心遭遇阻力. 标普因AI资本支出风险下调甲骨文评级,而地方抗议和纽约对超大规模数据中心暂停许可反映出对AI基础设施抵制的加剧。
- S&P Global sees OpenAI as a "key credit risk" for Oracle and cuts its credit rating(在新标签页中打开)
- New York bans data center construction for a year, rattling AI industry(在新标签页中打开)
- New York State halts construction of all new data centers(在新标签页中打开)
- The fight against AI data centers is just beginning(在新标签页中打开)
AI融资激增. Databricks以1880亿美元估值融资,Meta正与Anthropic谈判100亿美元的数据中心租赁,Nous Research为其开源智能体获得7500万美元。
能源与硬件赌注. 能源公司自互联网时代以来筹集了最多资金为AI供电,一笔由SambaNova芯片支持的4亿美元贷款表明对非GPU AI硬件的投资正在增加。
设备端AI突破
1比特模型登陆手机. PrismML的Bonsai将Qwen3.6-27B压缩至3.9GB,同时保留90%的基准测试分数,实现了设备端工具调用智能体;据悉苹果正在谈判。
- Bonsai 27B: The First 27B-Class Model to Run on a Phone(在新标签页中打开)
- Bonsai 27B: 1-bit dense LLM running locally in your browser using custom WebGPU kernels(在新标签页中打开)
- Prism-ML Bonsai Qwen 3.6 27B(在新标签页中打开)
- PrismML’s new Ternary Qwen3.6 27B runs near fp16 precision on 10GB of memory!!!(在新标签页中打开)
- So what's the consensus on 1bit models? Is it still a pipe dream?(在新标签页中打开)
- PrismML Bonsai 27B is surprisingly usable on the Jetson Orin Nano 8GB(在新标签页中打开)
- Is anyone having any luck with the Ternary Bonsai 27B DFlash?(在新标签页中打开)
- Can we get a "not base model" flair?(在新标签页中打开)
- Bonsai 27B runs locally on an iPhone - a 27B model in 3.9GB(在新标签页中打开)
- User experience of Bonsai-Ternary-27B on 4060Ti 16GB for KB management and productivity assistant use cases(在新标签页中打开)
- Bonsai 27B is a full open reasoning model that fits on an iPhone(在新标签页中打开)
- Bonsai-27B & Ternary-Bonsai-27B - Updates (on PRs)(在新标签页中打开)
- Apple in talks with startup PrismML that shrinks AI models to run on an iPhone(在新标签页中打开)
家庭实验室运行前沿模型. llama.cpp的优化如DFlash将MoE模型加速最高6倍,一加手机从闪存流式传输专家以1.3 tok/s的速度运行60GB模型。
- I tested all llama.cpp's speculative decoding methods on Qwen 3.6 27B: MTP ~2.7x, DFlash ~3.7x, n-gram stack ~6x on real coding. Local AI win. My findings on RTX 6000 PRO.(在新标签页中打开)
- DFlash makes Qwen3.6 27B 2.2x faster with no quality loss(在新标签页中打开)
- GPT-OSS-120B, Qwen 30B and Gemma 26B on an Android phone at 1-5 tok/s: +60GB model, 11GB of RAM, CPU only(在新标签页中打开)
政策、诉讼与开源对峙
苹果与OpenAI法律战. 苹果起诉OpenAI窃取商业机密,指控其挖走超过400名员工,包括首席硬件官,威胁到OpenAI的IPO和云信任。
- Apple sues OpenAI after ex-engineer allegedly used bug to steal trade secrets(在新标签页中打开)
- The wildest allegations in Apple’s trade secrets lawsuit against OpenAI(在新标签页中打开)
- The 6 wildest claims in Apple’s lawsuit against OpenAI(在新标签页中打开)
- OpenAI pushes back on Apple trade secret lawsuit(在新标签页中打开)
- Sam Altman didn’t need another lawsuit(在新标签页中打开)
- How Apple’s big lawsuit could disrupt OpenAI’s IPO plans(在新标签页中打开)
- Apple’s plot to crush OpenAI(在新标签页中打开)
- Apple’s lawsuit couldn’t come at a worse time for OpenAI(在新标签页中打开)
中国开源浪潮. 中国模型现在占HuggingFace下载量的41%,Kimi K3可与西方领先者竞争,中国启动了排除西方的并行AI治理机构。
- The real AI race may no longer be at the frontier(在新标签页中打开)
- Source: the Trump administration and industry groups discussed streamlining US open model releases of equal or lesser capability to leading Chinese open models(在新标签页中打开)
- Chinese President Xi Jinping speaks at World AI Conference and reaffirms commitment to open source to promote"openness and win-win"(在新标签页中打开)
- China’s Xi Touts Open-Source AI and Takes a Swipe at U.S. Dominance(在新标签页中打开)
- China's new World Artificial Intelligence Cooperation Organization is President Xi's clearest play yet for a parallel AI order(在新标签页中打开)
问责呼声日益高涨. 德国监管机构认定聊天机器人为内容负责,xAI起诉一名用户生成CSAM,Demis Hassabis提议成立类似FINRA的机构进行前沿安全审查。
这是本周回顾 – 下周日见。