报告形态 · 五维分类 + 风险 + 行动 + 时间线 ← 回站点
WEEK 2026-09-17 · 46 STORIES · SKILLHUB WEEKLY

SkillHub.news 每周AI

9 月 11 日 ~ 2026 年 9 月 17 日 · 周报 · 素材 4
本期看点46 条 · 约18分钟
01
本周主题线WEEKLY THREADS24 条
FOMC决议落地:12:0加息25bp至3.75%-4.00%
02
官方事件精选OFFICIAL EVENTS14 条
smartARM: The AI-Powered Bionic Arm
03
榜单异动RANKING MOVES8 条
obra/superpowers 星标 +79,301
01本周主题线WEEKLY THREADS24 条

FOMC决议落地:12:0加息25bp至3.75%-4.00%

央视/每经/中证报多源2026-09-17置信度 高
今晨决议:2023年7月来首次加息、降息周期终结;点阵图4.1%、16/18官员年内再加、核心PCE上修3.4%、2%推迟2029、沃什偏鹰、CME 10月再加息53%;美股整体跳水(道指-1.21%)、10Y升至5.008%——"高利率更久"成基准情形(央视/每经/中证报多源,高)

OpenRouter周榜易主(进展·第2天)

OpenRouter官方2026-09-17置信度 高进展·第 2 天
截至9/13周报:GPT-5.6 Luna 18.20万亿(+41.1%)登顶、美系自4月下旬约20周来首回周Token榜首、终结中国近两月轮转;Hy4退第二16.80万亿、DeepSeek厂商第一21.76万亿;总量127万亿为已报口径;实时榜(9/15)窗口不同数字不同,引用必带窗口(OpenRouter官方,高)

Gemini 3.8 Live/ET双语音模型

Google官方+多源2026-09-17置信度 高
9/15发布:ET版82.6登顶AA语音对语音质量榜(超GPT-Live-1的81.5,AA注明单次试验口径)、标准版实测$0.84/小时(ET版$3.50)、语音中并行后台调用工具、97语言——语音成中美座次决战区(Google官方+多源,高)

Qwen3.8-27B登顶HF史上Most-likes

HF+AIBase/网易多源2026-09-17置信度 高
超FLUX.1/R1/K3/Llama-3成为平台历史点赞第一,社区口碑首超跑分(8/14开源约一个月,勿写"不足一月");Perplexity选为本地Agent模型;阿里累计开源460+/下载30亿+(HF+AIBase/网易多源,高)

美两党《Stop Rogue AI Act》

报道台账2026-09-17
9/9提出/9/15官网:NIST一年内定智能体标准(机器可读清单/可验证身份/防篡改日志/行为监测),联邦承包商经FAR强制——首部聚焦"行为管控"而非能力上限的两党联邦立法,关联7月700智能体攻破HF事件(议员官网,内容高/日期中)

香港施政报告AI治理七方面

官方121-123条核到2026-09-17置信度 高
9/16:政务司长统筹AI风险治理+数字办设AI专员;法改会研深伪立法;财库局年内就下架AI生成金融诈骗内容立法咨询;律政司研AI致损责任;明年发布全社会AI Agent安全管理指引——大中华区首个系统性Agent治理样本(官方121-123条核到,高)

字节AI制药分拆Anew Labs首融

报道台账2026-09-17
2.9亿美元/估值15亿/HSG+IDG+高瓴领投、五源联合领投、字节保留56%——"大厂分拆+独立融资"成AI变现新通道(Reuters 9/16+多源,中高,匿名信源官方未确认)

超衍智能RSI近4亿元

央广网/科创板日报2026-09-17置信度 高
清华95后陈勇超6月创办,天使+天使+累计近4亿(天使轮IDG+星连+晶泰领投;天使+轮深创投/中关村科学城/上海未来产业基金)——递归自我改进(RSI)获国资+VC重注,追赶逻辑从追前沿改为改进速度(央广网/科创板日报,高)

优必选U1交付剪刀差(进展·第4天)

财报电话会/南方+2026-09-17置信度 中高进展·第 4 天
9/16深圳首批交付(Lite 11.98万起)但电话会披露年内可交付仅1500-2000台 vs 订单13361台,剪刀差首次官方量化;锚定2026Q4单季EBITDA"有机会转正";⚠️台账勘误:9/14条目H1出货应为16123台(财报电话会/南方+,中高)

Mozilla开源AI报告v1.1

Mozilla/Ars2026-09-17置信度 高
9/15发布:开源权重与闭源差距仅4.4个月(METR拟合,Kimi K3比Fable 5低3分成本30%);OpenRouter 8月Top10中8个开放权重(7个中国产)但收入仅4%(2025年5-9月窗口口径)——闭源溢价仅存8-12小时任务区间(Mozilla/Ars,高)

康惠股份17.2亿算力长约

公告原文9/17刊载2026-09-17置信度 高
五年算力服务合同17.2亿(含税)+11.41亿服务器购销锁定供给;负债率69%→78%、2026新增收入仅约3000万——算力长单下沉中小上市公司,"重资产换长约"现金流风险需警惕(公告原文9/17刊载,高)

资本快照三连:Euclyd/TypeSafe/CADDi

报道台账2026-09-17
Euclyd $231M A轮(三星联合领投/前ASML CEO Wennink任董事长,非GPU推理架构,"100倍能效"系公司预测);TypeSafe $40M种子(DCVC领投,决策模型Jev不生成文本,快193倍系公司自测无独立评测);CADDi $1.14亿D轮估值12亿(丰田Woven/Salesforce Ventures参投,制造业图纸数据)(CNBC/DCVC/Fortune+WSJ,融资高/宣称低)

一句话带过/背景存档

企业自报2026-09-17置信度 中
王室AI峰会本周邓弗里斯+Ditchley"共同原则"宪章草案(中高);加拿大Float:企业AI支出两年12倍/47%渗透/户均2.71供应商(真实流水非普查,中高);华为CANN开源社区月活6230(企业自报,中);NPCI支付Agent注册表(9/10首发超48h窗口降背景,"SAFR竞速"子项未证实已删)

剔除项存档

报道台账2026-09-17
地平线征程1500万颗里程碑系9/3已官宣(9/15仅补办仪式)超时效剔除;欧盟AI法高风险首轮检查仅聚合简报源、无官方公告暂缓;NPCI注册表超48h窗口降背景;阶跃StepAudio 3为9/16已报口径不重复

OpenAI被曝1.2万亿美元IPO前融资

FT2026-09-16置信度 高
FT 9/15:与大型投资者商谈IPO前私募融资,目标估值1.2万亿(较3月8520亿投后+40%)、由投资者发起、早期阶段;数见智核验至FT一手(非The Information转述);与Altman"2026不上市"不矛盾——恰是推迟背景下的抢筹逻辑(FT,高)

FOMC前夜紧缩定价拉满(进展·第2天)

财联社/新华社2026-09-16置信度 高进展·第 2 天
9/15美盘10Y盘中5.04%(口径上修"2007年来最高")、CME加息25bp定价92%、20家投行17家押注;WTI结算105.83(108.59系误记)、柴油期货历史新高;今夜零售销售(预期+0.9%)+明晨决议同日落(财联社/新华社,高)

评测双向信任危机:假阴性侧实锤(进展)

arXiv/ARC Prize2026-09-16置信度 高进展
arXiv 2609.13009(51专家重评6大物理基准,预印本未同行评审):多数"错误作答"实为判分器/参考答案错误——HLE-Physics 47.3%→78.7%(保留子集口径)、CMT 61.0→87.2、PRISM 13.0→94.6;叠加ARC-AGI-3 harness差距62.7% vs 99.9%官方敲实;ARC-AGI-4明年Q1/ARC-AGI-5"发明"立项(arXiv/ARC Prize,高)

国产"小快省"三连发

多源2026-09-16置信度 高
阶跃StepAudio 3五款语音模型登顶AA语音榜(Realtime 98.9%对话动态第一/ASR WER 1.7%并列第一);云知声U2-Flash 266B总参/约10B激活、Agent迭代步数与Token降20-30%(港交所公告一手,DeepSWE v1.1 64.6);紫东太初5.0-9B空间理解开源(9基准8项10B内第一)(多源,高)

政企AI采购下沉连接层:MCP代理首入采购清单

官方公示2026-09-16置信度 高
中原银行87万"企业级AI网关"中标候选公示(采购清单首次出现MCP代理/AI Agent代理)+国寿养老86万智能体集群管控平台+同窗50-320万政企AI单密集落地——AI网关复制当年API网关独立品类路径(官方公示,高)

Factory 2亿美元融资估值50亿

Reuters2026-09-16置信度 高
4月15亿→50亿约7个月3.3倍,Blackstone/Khosla/红杉/Insight参投——继Cognition 480亿后PE巨头把编码Agent当基础设施定价(Reuters,高)

Profound 1.8亿美元D轮估值18亿

官方通稿2026-09-16置信度 高
AEO/GEO赛道首个十亿级轮次,红杉+Kleiner Perkins联合领投,服务1000+企业品牌(财富100强约1/3),距上轮9600万C轮不足7个月(官方通稿,高)

DeepSeek拟聘任首任CFO(进展)

Reuters/36氪2026-09-16置信度 中进展
Reuters 9/14独立报道:拟聘高榕合伙人严文涛(1991年生,投过MiniMax/智谱)出任首任CFO——科创板线(进展)新增制度化信号;plans to hire口径,本人与公司均未确认(Reuters/36氪,中)

中国智驾供应链出海双单

证券时报2026-09-16置信度 高
禾赛获美国头部Robotaxi公司超4000万美元激光雷达独家订单(唯一供应商、2026年底交付);均胜电子新获两家头部主机厂智能化定点、全生命周期约150亿元(证券时报,高)

Waymo双城记

Reuters2026-09-16置信度 高
拉斯维加斯首次向公众开放全无人服务(第六代Driver/极氪平台);联手日本GO/日之交通目标2027年东京上线日本首个全无人商用(Reuters,高)
02官方事件精选OFFICIAL EVENTS14 条

smartARM: The AI-Powered Bionic Arm

YouTube·Meta2026-09-17置信度 高
原文 ↗

Middleware for Managed Deep Agents

YouTube·LangChain2026-09-17置信度 高
原文 ↗

Reception by ElevenAgents

YouTube·ElevenLabs2026-09-17置信度 高
原文 ↗

Quoting Mustafa Suleyman

官方·Simon Willison2026-09-17置信度 高
原文 ↗

Advance your career. Advance the world.

YouTube·AMD2026-09-17置信度 高
原文 ↗

See How The Grout Guy Scales Service with Agentforce

YouTube·Salesforce2026-09-17置信度 高
原文 ↗

NVIDIA HGX B300: on-site with CoreWeave engineers

YouTube·Weights & Biases2026-09-17置信度 高
原文 ↗

Claude Cowork and chat are now one Claude

官方·Simon Willison2026-09-17置信度 高
原文 ↗
03榜单异动RANKING MOVES8 条

obra/superpowers 星标 +79,301

GitHub2026-09-03 → 2026-09-10
原文 ↗

mattpocock/skills 星标 +71,905

GitHub2026-09-03 → 2026-09-10
原文 ↗

anthropics/skills 星标 +49,058

GitHub2026-09-03 → 2026-09-10
原文 ↗

farion1231/cc-switch 星标 +36,887

GitHub2026-09-03 → 2026-09-10
原文 ↗
生态脉搏站点数据自更新
2026-09-03 → 2026-09-10 星标涨幅:obra/superpowers +79,301★ · mattpocock/skills +71,905★ · multica-ai/andrej-karpathy-skills +59,162★ · anthropics/skills +49,058★ · Shubhamsaboo/awesome-llm-apps +38,266★ · farion1231/cc-switch +36,887★ · nextlevelbuilder/ui-ux-pro-max-skill +35,298★ · Graphify-Labs/graphify +32,474★
※风险分级灰犀牛 / 黑天鹅
灰犀牛 · 可预见
  • G1 高利率更久刺破 AI 估值尾部:点阵图 16/18 位官员预计年内再加、核心 PCE 上修至 3.4%;若 10/11 月核心 PCE 不降反升、10Y 站稳 5% 上方,纯叙事型 AI(无收入开源项目、烧钱交付的机器人厂商)将遭再融资与估值双杀(高)
  • G2 人形机器人「订单-交付」信任折价:13361 台订单 vs 1500-2000 台交付已官方量化,若 2026Q4 EBITDA 转正跳票,全行业面临「订单真实性」集体重估;实时仪表盘 = 优必选月度交付数据(中高)
  • G3 智能体合规成本成中小玩家出清器:NIST 一年内标准 + 联邦承包商强制 + 欧盟首轮检查启动,合规是固定成本,中小 Agent 厂商将被迫退出或依附大厂(中)
黑天鹅 · 低概率大冲击
  • B1 OpenAI vs Apple 听证会(10/1)出现「平台封禁」级判决:若反垄断救济含排他性分发限制,直接改写模型分发格局(中)
  • B2 RSI 失控性能力跳变:「AI 做研究」重注若产生不可解释的能力跃升,全球监管将从「行为管控」急转为「能力冻结」——训练暂停、资本恐慌(低-中)
  • B3 Anthropic IPO 路演失败或估值腰斩(10 月中旬):AI 第一权重股定价若失败,整个一级市场退出通道冻结(低-中)
✦行动建议3
1. 智能体合规审计:标准定型期做「Agent 三件套」

面向出海 AI 公司与接政企单的团队,提供「智能体身份标识 + 防篡改操作日志 + 行为审计」三件套 SDK / 中间件与合规自查服务。路径:拆 NIST / 香港 / NPCI 三份文件做中文对照 checklist(免费报告)→ 2-3 家灯塔客户交付轻量审计中间件 → 沉淀标准产品,绑定「联邦承包商强制」出海链。

时间窗 6-12 个月(NIST 标准制定期 = 对齐期,标准发布后即成红海)
2. 语音原生垂直 Agent:吃 $0.84/小时 的底座红利

不做底座做交付——中小商家 AI 电话接单 / 预约前台(餐饮、家政、诊所)、AI 语音陪练,按结果收费。路径:用国产低价底座一周内做「餐饮订位电话 Agent」demo → 3 家试点验证接通率与成交率 → 按「接一单 X 元」复制到 50 家 → 叠加并行工具调用直连订单系统。

时间窗 12-18 个月(大厂亲自下场做垂直 SaaS 前)
3. 开源变现缺口:私有化部署 + Token 成本路由

吃「开源占 Token 8 成、收入仅 4%」的缺口——给数据敏感中小企业做 Qwen3.8-27B 级私有化部署 + 行业微调;给高消耗团队做开源混合路由,砍 30-50% 推理账单。路径:发布《开源 vs 闭源选型指南 + 成本测算表》→ 接 5 个私有化单(27B 单卡可跑)→ 沉淀法律 / 电商 / 教育行业模板转订阅 + 路由抽成。

时间窗 约 12 个月(差距收窄到 3 个月以内前话术最强)
⏱未来 3 个月时间线概率
时间窗事件概率依据
9/24长江证券智能体项目截标高券商 AI 采购同业复制期验证点,中标价决定后续银行跟随节奏
9/29OpenAI DevDay高Agent 产品化与合规 API 布局将直接冲击 Token 榜座次
9/29上海公安 1212.5 万智能体平台开标高警务级 Agent 平台首标落定,政务 AI 采购风向标
9/30加州 AI 法案签署窗口截止 + ARC-AGI-3 截榜高州级立法与联邦形成双层合规压力;截榜检验 4.4 个月差距是否继续收窄
10/1OpenAI vs Apple 听证会开庭高(出结果低)分发层反垄断攻防首回合风向标
10 月上中旬10 月 CPI高核心 PCE 上修后首个验证窗:不降则 10 月加息固化,G1 逼近
10 月中旬Anthropic IPO 路演中(60%)AI 一级市场定价锚:成功则资本新范式获二级验证,失败触发 B3
10 月 FOMC加息决断中(50-60%)测试「利空出尽」修复能否二次成立,费半是先行指标
10 月底SaaS 三季报高AI 收入是否撑得起高利率估值的试金石
11 月OpenRouter 月度报告高验证 Luna 登顶是趋势还是单周波动;月环比转负是需求侧首次预警
12 月 FOMC年内第三次加息决断中(40-50%)兑现点阵图则 2027 降息预期重定价;不加则紧缩见顶交易开启
12 月(推断)优必选 Q4 EBITDA 转正验证窗中(40%)G2 剪刀差是否收敛在此见分晓
12 月(推断)开源-闭源差距更新中(55%)收窄则追赶范式强化;走阔则 RSI 投入开始产生代差
素材来源:报道台账(滚动查重):2026-09-11 ~ 2026-09-17 共 99 条主题记录 · AI 官方视频监控台账:7 天 · 165 条事件(精选 14 条带链接) · 站点榜单快照:2026-09-03 → 2026-09-10 星标涨幅 Top8 · 风险/行动/时间线:承接 2026-09-17 日报
SkillHub.news — 全球 Skill 榜单
WEEK 2026-09-17 · 46 STORIES · SKILLHUB WEEKLY
SkillHub.news Weekly AI
Sep 11 ~ Sep 17, 2026 · Weekly · sources 4
Highlights46 stories · approx.18min read
01
WEEKLY THREADS24 stories
FOMC decision lands: 12:0 vote to hike 25bp to 3.75%-4.00%
02
OFFICIAL EVENTS14 stories
smartARM: The AI-Powered Bionic Arm
03
RANKING MOVES8 stories
obra/superpowers stars +79,301
01WEEKLY THREADS24 stories

FOMC decision lands: 12:0 vote to hike 25bp to 3.75%-4.00%

央视/每经/中证报多源2026-09-17Confidence High
This morning's decision: first hike since July 2023, ending the rate-cut cycle; dot plot 4.1%, 16/18 officials to hike again within the year, core PCE revised up to 3.4%, 2% delayed to 2029, Warsh leans hawkish, CME 53% for another hike in October; U.S. stocks broadly plunge (Dow -1.21%), 10Y rises to 5.008% — 'higher rates for longer' becomes the baseline case (CCTV/National Business Daily/China Securities Journal, multiple sources, high)

OpenRouter weekly leaderboard changes hands (progress · Day 2)

OpenRouter官方2026-09-17Confidence HighProgress · Day 2
As of the 9/13 weekly report: GPT-5.6 Luna tops at 18.20 trillion (+41.1%), U.S. models return to the weekly Token top spot for the first time in about 20 weeks since late April, ending nearly two months of Chinese rotation; Hy4 drops to second at 16.80 trillion, DeepSeek first among vendors at 21.76 trillion; total 127 trillion on a reported basis; real-time leaderboard (9/15) varies by window, citations must include the window (OpenRouter official, high)

Gemini 3.8 Live/ET dual speech models

Google官方+多源2026-09-17Confidence High
9/15 release: ET version scores 82.6 to top AA speech-to-speech quality leaderboard (surpassing GPT-Live-1's 81.5; AA notes single-trial basis), standard version measured at $0.84/hour (ET version $3.50), parallel background tool calls during speech, 97 languages—voice becomes a decisive battleground for China-U.S. rankings (Google official + multiple sources, high)

Qwen3.8-27B tops HF's all-time Most-likes

HF+AIBase/网易多源2026-09-17Confidence High
Surpasses FLUX.1/R1/K3/Llama-3 to become platform's all-time No. 1 in likes; community reputation surpasses benchmark scores for the first time (open-sourced on 8/14, about one month ago); selected by Perplexity as local Agent model; Alibaba cumulative open-source 460+, downloads 3 billion+ (HF + AIBase/NetEase multiple sources, high)

US bipartisan Stop Rogue AI Act

报道台账2026-09-17
Introduced 9/9; official website 9/15: NIST to set agent standards within one year (machine-readable inventory/verifiable identity/tamper-proof logs/behavior monitoring), mandatory for federal contractors via FAR—the first bipartisan federal legislation focused on "behavior control" rather than capability caps; linked to July incident in which 700 agents breached HF (lawmaker website, content high/date medium)

Hong Kong Policy Address: seven areas of AI governance

官方121-123条核到2026-09-17Confidence High
9/16: Chief Secretary for Administration to coordinate AI risk governance + Digital Policy Office to establish an AI Commissioner; Law Reform Commission to study deepfake legislation; Financial Services and the Treasury Bureau to consult within the year on legislation for taking down AI-generated financial scam content; Department of Justice to study liability for AI-caused harm; next year to issue society-wide AI Agent safety management guidelines—Greater China's first systematic Agent governance sample (official paras. 121-123 verified, high)

ByteDance AI pharma spin-off Anew Labs raises first round

报道台账2026-09-17
$290M / $1.5B valuation / HSG + IDG + Hillhouse led, 5Y Capital co-led, ByteDance retains 56% — 'Big Tech spin-off + independent financing' becomes a new path for AI monetization (Reuters 9/16 + multiple sources, medium-high, anonymous sources, not officially confirmed)

Chaoyan Intelligence RSI nearly 400 million yuan

央广网/科创板日报2026-09-17Confidence High
Founded in June by Tsinghua post-95s Chen Yongchao; Angel + Angel+ cumulative nearly 400 million (Angel round led by IDG, Xinglian, XtalPi; Angel+ round by Shenzhen Capital Group / Zhongguancun Science City / Shanghai Future Industry Fund) — recursive self-improvement (RSI) gets heavy bets from state-owned capital + VC, with the catch-up logic shifting from chasing the frontier to improving speed (CNR / STAR Market Daily, High)

优必选U1交付剪刀差(进展·第4天)

财报电话会/南方+2026-09-17Confidence MidProgress · Day 4
9/16深圳首批交付(Lite 11.98万起)但电话会披露年内可交付仅1500-2000台 vs 订单13361台,剪刀差首次官方量化;锚定2026Q4单季EBITDA"有机会转正";⚠️台账勘误:9/14条目H1出货应为16123台(财报电话会/南方+,中高)

Mozilla Open Source AI Report v1.1

Mozilla/Ars2026-09-17Confidence High
9/15 release: open-weight vs. closed-source gap only 4.4 months (METR fit; Kimi K3 3 points below Fable 5, 30% cost); 8 of OpenRouter's August Top10 are open-weight (7 Chinese-made) but only 4% of revenue (2025 May-September window); closed-source premium exists only in the 8-12 hour task range (Mozilla/Ars, High)

Kanghui Co.'s 17.2亿 compute long-term contract

公告原文9/17刊载2026-09-17Confidence High
Five-year compute service contract 17.2亿 (tax included) + 11.41亿 server purchase/sale locks in supply; debt ratio 69%→78%, 2026 new revenue only about 3000万—long-term compute orders moving down to small/mid-cap listed companies, cash-flow risk of 'heavy assets for long-term contracts' warrants caution (original announcement published 9/17, High)

Capital snapshot trio: Euclyd/TypeSafe/CADDi

报道台账2026-09-17
Euclyd $231M Series A (Samsung co-led/former ASML CEO Wennink as chairman, non-GPU inference architecture, '100倍 energy efficiency' is company projection); TypeSafe $40M seed (DCVC led, decision model Jev does not generate text, 193倍 faster is company self-test with no independent evaluation); CADDi $1.14亿 Series D at 12亿 valuation (Toyota Woven/Salesforce Ventures participated, manufacturing drawing data) (CNBC/DCVC/Fortune+WSJ, funding high/claims low)

Brief mention/background archive

企业自报2026-09-17Confidence Mid
Royal AI Summit this week: Dumfries + Ditchley "common principles" charter draft (medium-high); Canada Float: enterprise AI spending 12x in two years/47% penetration/average 2.71 vendors per account (real transaction data, not a census, medium-high); Huawei CANN open-source community MAU 6230 (company self-reported, medium); NPCI payment Agent registry (9/10 first release; beyond 48h window, downgraded to background; "SAFR race" sub-item unverified and deleted)

Excluded items archive

报道台账2026-09-17
Horizon Journey 15-million-unit milestone was officially announced on 9/3 (9/15 was only a make-up ceremony); excluded as out of time window; EU AI Act high-risk first-round checks only from aggregated briefing sources, no official announcement, deferred; NPCI registry beyond 48h window, downgraded to background; StepFun StepAudio 3 is the 9/16 already-reported item, not repeated

OpenAI reportedly raising $1.2 trillion pre-IPO financing

FT2026-09-16Confidence High
FT 9/15: In talks with large investors for pre-IPO private financing; target valuation $1.2 trillion (+40% vs $852 billion post-money in March), investor-initiated, early stage; Shujianzhi verified to FT first-hand (not relayed by The Information); does not contradict Altman's "no IPO in 2026"—it is precisely the scramble-for-allocation logic amid delay (FT, High)

FOMC Eve: Tightening Pricing Maxed Out (Progress · Day 2)

财联社/新华社2026-09-16Confidence HighProgress · Day 2
9/15 US session: 10Y hit 5.04% intraday (measure revised up to "highest since 2007"), CME prices 92% for a 25bp hike, 17 of 20 investment banks bet on it; WTI settled at 105.83 (108.59 was misrecorded), diesel futures hit a record high; tonight's retail sales (expected +0.9%) + tomorrow morning's decision both land on the same day (Cailian Press/Xinhua, high)

Evaluation's two-way trust crisis: false-negative side confirmed (update)

arXiv/ARC Prize2026-09-16Confidence HighProgress
arXiv 2609.13009 (51 experts re-evaluate 6 major physics benchmarks; preprint not peer-reviewed): most "incorrect answers" were actually scorer/reference-answer errors—HLE-Physics 47.3%→78.7% (retained-subset basis), CMT 61.0→87.2, PRISM 13.0→94.6; compounded by ARC-AGI-3 harness gap of 62.7% vs 99.9%, officially confirmed; ARC-AGI-4 next year Q1 / ARC-AGI-5 "invention" project initiated (arXiv/ARC Prize, high).

Domestic "small, fast, cheap" trio launches

多源2026-09-16Confidence High
StepFun StepAudio 3's five speech models top the AA speech leaderboard (Realtime 98.9% dialogue dynamics first / ASR WER 1.7% tied for first); Unisound U2-Flash 266B total params / ~10B active, Agent iteration steps and Token down 20-30% (first-hand from HKEX announcement, DeepSWE v1.1 64.6); Zidong Taichu 5.0-9B spatial understanding open-sourced (first within 10B on 8 of 9 benchmarks) (multi-source, high).

Government-enterprise AI procurement moves down to the connectivity layer: MCP proxy first enters procurement list

官方公示2026-09-16Confidence High
Zhongyuan Bank 87万 "enterprise-grade AI gateway" winning-bid candidate public notice (MCP proxy/AI Agent proxy appear in procurement list for the first time) + China Life Pension 86万 agent cluster management and control platform + Tongchuang 50-320万 government-enterprise AI orders see intensive rollout — AI gateway replicates the path of API gateway as an independent category back then (official announcement, high)

Factory raises $200M at $5B valuation

Reuters2026-09-16Confidence High
April $1.5B → $5B in about 7 months, 3.3x; Blackstone/Khosla/Sequoia/Insight participated—after Cognition's $48B, PE giants price coding agents as infrastructure (Reuters, High)

Profound's $180M Series D at a $1.8B valuation

官方通稿2026-09-16Confidence High
AEO/GEO sector's first billion-scale round, co-led by Sequoia + Kleiner Perkins, serving 1000+ enterprise brands (about 1/3 of Fortune 100), less than 7 months after the previous $96M Series C (official press release, high)

DeepSeek plans to appoint first CFO (progress)

Reuters/36氪2026-09-16Confidence MidProgress
Reuters 9/14 independent report: plans to hire Gaorong partner Yan Wentao (born 1991; backed MiniMax/Zhipu) as first CFO—STAR Market track (progress) adds institutionalization signal; wording is "plans to hire"; neither he nor the company has confirmed (Reuters/36Kr, Medium)

Chinese smart-driving supply chain goes overseas with two orders

证券时报2026-09-16Confidence High
Hesai wins an exclusive lidar order worth over $40 million from a leading U.S. Robotaxi company (sole supplier, delivery by end-2026); Joyson Electronics secures new intelligent-tech design wins from two leading OEMs, worth about RMB 15 billion over the full lifecycle (Securities Times, high)

Waymo's Tale of Two Cities

Reuters2026-09-16Confidence High
Las Vegas opens fully driverless service to the public for the first time (sixth-generation Driver/Zeekr platform); partners with Japan's GO and Nihon Kotsu, aiming to launch Japan's first fully driverless commercial service in Tokyo in 2027 (Reuters, High)
02OFFICIAL EVENTS14 stories

smartARM: The AI-Powered Bionic Arm

YouTube·Meta2026-09-17Confidence High
Original ↗

Middleware for Managed Deep Agents

YouTube·LangChain2026-09-17Confidence High
Original ↗

Reception by ElevenAgents

YouTube·ElevenLabs2026-09-17Confidence High
Original ↗

Quoting Mustafa Suleyman

官方·Simon Willison2026-09-17Confidence High
Original ↗

Advance your career. Advance the world.

YouTube·AMD2026-09-17Confidence High
Original ↗

See How The Grout Guy Scales Service with Agentforce

YouTube·Salesforce2026-09-17Confidence High
Original ↗

NVIDIA HGX B300: on-site with CoreWeave engineers

YouTube·Weights & Biases2026-09-17Confidence High
Original ↗

Claude Cowork and chat are now one Claude

官方·Simon Willison2026-09-17Confidence High
Original ↗
03RANKING MOVES8 stories

obra/superpowers stars +79,301

GitHub2026-09-03 → 2026-09-10
Original ↗

mattpocock/skills stars +71,905

GitHub2026-09-03 → 2026-09-10
Original ↗

anthropics/skills stars +49,058

GitHub2026-09-03 → 2026-09-10
Original ↗

farion1231/cc-switch stars +36,887

GitHub2026-09-03 → 2026-09-10
Original ↗
Ecosystem Pulseself-updating
2026-09-03 → 2026-09-10 Star growth:obra/superpowers +79,301★ · mattpocock/skills +71,905★ · multica-ai/andrej-karpathy-skills +59,162★ · anthropics/skills +49,058★ · Shubhamsaboo/awesome-llm-apps +38,266★ · farion1231/cc-switch +36,887★ · nextlevelbuilder/ui-ux-pro-max-skill +35,298★ · Graphify-Labs/graphify +32,474★
※Risk Levelsgray rhino / black swan
Gray rhino · foreseeable
  • G1 Higher-for-longer rates puncture the AI valuation tail: dot plot shows 16/18 officials expect another hike this year, core PCE revised up to 3.4%; if core PCE for 10/11 rises instead of falls and the 10Y holds firmly above 5%, pure-narrative AI (no-revenue open-source projects, robot makers burning cash on delivery) will face a double hit from refinancing and valuation (High)
  • G2 Humanoid robot 'order-delivery' trust discount: 13361-unit orders vs 1500-2000-unit deliveries officially quantified; if the 2026Q4 EBITDA positive turn is missed, the entire industry faces a collective re-rating of 'order authenticity'; real-time dashboard = UBTech monthly delivery data (Medium-High)
  • G3 Agent compliance costs become a shakeout mechanism for small and medium players: NIST standards within one year + federal contractor mandates + EU first-round inspections begin; compliance is a fixed cost, and small and medium Agent vendors will be forced to exit or attach to big tech (Medium)
Black swan · low-probability
  • B1 OpenAI vs Apple hearing (10/1) produces a 'platform ban'-level ruling: if antitrust remedies include exclusive distribution restrictions, it directly rewrites the model distribution landscape (Medium)
  • B2 RSI runaway capability leap: If heavy bets on "AI doing research" produce an inexplicable capability jump, global regulators will abruptly pivot from "behavior control" to "capability freeze"—training pauses, capital panic (low-medium)
  • B3 Anthropic IPO roadshow fails or valuation is halved (mid-October): If pricing for the top-weighted AI stock fails, the entire primary market exit channel freezes (low-medium)
✦Action Items3
1. Agent compliance audit: build the 'Agent three-piece set' during the standard finalization period

For AI companies going overseas and teams taking government/enterprise contracts, provide a three-piece SDK / middleware set and compliance self-check service covering 'Agent identity + tamper-proof operation logs + behavior auditing.' Path: break down the three documents NIST / Hong Kong / NPCI into a Chinese comparison checklist (free report) → deliver lightweight audit middleware to 2-3 lighthouse customers → refine into a standard product, bound to the 'federal contractor mandate' overseas expansion chain.

window 6-12 months (NIST standard-setting period = alignment period; once the standard is released, it becomes a red ocean)
2. Voice-native vertical Agent: capturing the $0.84/hour base-layer dividend

Don’t build the base layer—build delivery: AI phone order-taking / appointment front desk for SMBs (restaurants, home services, clinics), AI voice practice partner, charged on outcomes. Path: use a low-cost domestic base layer to build a Restaurant Reservation Phone Agent demo within one week → validate connection and conversion rates with 3 pilot merchants → replicate to 50 merchants on a per-order X yuan basis → add parallel tool calling to connect directly to order systems.

window 12-18 months (before big tech enters vertical SaaS themselves)
3. Open-source monetization gap: private deployment + Token cost routing

Capitalize on the gap where open source accounts for 80% of Tokens but only 4% of revenue—provide data-sensitive SMEs with Qwen3.8-27B-level private deployment + industry fine-tuning; provide high-consumption teams with open-source hybrid routing, cutting inference bills by 30-50%. Path: publish an Open Source vs Closed Source Selection Guide + Cost Calculation Sheet → take 5 private deployment orders (27B can run on a single card) → turn legal / e-commerce / education industry templates into subscriptions + routing commissions.

window About 12 months (pitch is strongest before the gap narrows to within 3 months)
⏱Next 3 MonthsProb.
WindowEventProb.Basis
9/24Changjiang Securities AI agent project bidding closesHighBrokerage AI procurement peer-replication period validation point: winning bid price determines the pace at which banks follow
9/29OpenAI DevDayHighAgent productization and compliant API buildout will directly impact Token leaderboard rankings
9/29Shanghai police 1212.5 万 agent platform bid openingHighFirst police-grade Agent platform tender awarded; a bellwether for government AI procurement
9/30California AI bill signing window closes + ARC-AGI-3 leaderboard cutoffHighState-level legislation and federal government create dual-layer compliance pressure; leaderboard cutoff tests whether the 4.4-month gap continues to narrow
10/1OpenAI vs Apple hearing beginsHigh (low outcome)First-round bellwether in distribution-layer antitrust fight
10 月上中旬October CPIHighFirst validation window after core PCE upward revision: if it doesn't fall, October rate hike locks in; G1 nears
10 月中旬Anthropic IPO roadshowMedium (60%)AI primary-market pricing anchor: success would give the new capital paradigm secondary-market validation; failure triggers B3
10 月 FOMCRate-hike decisionMedium (50-60%)Testing whether the 'bad news exhausted' recovery can hold for a second time; SOX is the leading indicator
10 月底SaaS Q3 earningsHighA litmus test for whether AI revenue can support valuations under high interest rates
11 月OpenRouter monthly reportHighVerify whether Luna's rise to the top is a trend or a single-week fluctuation; a negative month-over-month turn is the first demand-side warning
12 月 FOMCThe year's third rate-hike decisionMedium (40-50%)If the dot plot is delivered, 2027 rate-cut expectations will be repriced; if no hike, the peak-tightening trade opens
12 月(推断)UBTech Q4 EBITDA positive-turn validation windowMedium (40%)Whether the G2 scissors gap converges will be decided here
12 月(推断)Open-source vs closed-source gap updateMedium (55%)If narrowing, the catch-up paradigm strengthens; if widening, RSI investment starts to create a generational gap
Sources:Coverage ledger (rolling deduplication):2026-09-11 ~ 2026-09-17, 99 topic records in total · AI Official Video Monitoring Log:7 days · 165 events (14 selected with links) · Site leaderboard snapshot:2026-09-03 → 2026-09-10 Top8 Star Growth · Risk/Action/Timeline:Follows the 2026-09-17 daily report
SkillHub.news — Global Skill rankings