AITOPAI热报

This week's charts

Ranked by heat score · 2026-09-21 – 2026-09-27
This weekWeek of 10-12Week of 10-05Week of 09-28Week of 09-21Week of 09-14Week of 09-07Week of 08-31Companies chart →Papers →

ASEAN top 10

10 stories
01
RoPA Manager Automates Vietnam Privacy Compliance Using Local LLMs
arXiv: DeepSeek· 09-24
Heat 93
02
泰国医院拥挤危机：6家初创企业用AI打造智慧医院生态
Techsauce· 09-21
Heat 82
03
T&B Media 与 Jaymart 在 ApsaraConference2026 分享 WonderClip 视频创作经验
阿里云 Alibaba Cloud· 09-25
Heat 81
04
新加坡内容公司 FizzDragon 在阿里云 Apsara 大会分享 AI 影视制作流程
阿里云 Alibaba Cloud· 09-25
Heat 81
05
李开复对话东南亚声音：开源与闭源模型之争关乎数据主权
李开复 Kai-Fu Lee· 09-24
Heat 81
06
MDEC CEO：数据中心投资只是起点，要让中小企业用上算力
MDEC· 09-24
Heat 81
07
VNPT 与 Qualcomm Technologies 扩大战略合作，推进边缘 AI 创新
Genk· 09-25
Heat 80
08
新加坡推出新一代AI翻译工具，专攻本地词汇与语境
联合早报· 09-26
Heat 79
09
Grab 与 OpenAI 推出 GO Forward with AI，帮助 30,000 名东南亚伙伴学习 AI 技能
OpenAI Blog· 09-23
Heat 77
10
Convai Innovations 在 Hugging Face 发布新模型卡
AI Will· 09-26
Heat 68

China top 10

10 stories
01
Xiaomi's MiMo-V2.6 RL training
OpenRouter· 09-21
Heat 128
02
Qwen-Image-2.1, a 7B-parameter open-source model for generation, editing, and transparency
shao__meng· 09-21
Heat 128
03
Xiaomi releases MiMo-V2.6-Pro model
Clement Delangue· 09-21
Heat 125
04
DeepSeek Details DSec Sandbox Platform for Agentic RL Training
pandaily· 09-24
Heat 124
05
Huawei Atlas 960E SuperPoD with 4,096 NPUs and 8 EFLOPS FP8
pandaily· 09-24
Heat 123
06
Huawei Launches OceanStor M900 Context Memory Storage for SuperPoDs
pandaily· 09-24
Heat 123
07
Alibaba Cloud Launches AgentCore, Next-Gen CPFS and HPN 8.0 Pro
pandaily· 09-23
Heat 123
08
Inspur MetaBrain SD200 Ultra Runs 2.8T Kimi K3 at 5.85ms/Token
pandaily· 09-23
Heat 123
09
China Telecom Releases Open-Source Xing4.0-29B-A4B Trained on Ascend
pandaily· 09-23
Heat 123
10
Xiaomi unveils HySparse2 architecture for MiMo-V3 agent inference
IT之家· 09-24
Heat 122

Global top 10

10 stories
01
Xiaomi's MiMo-V2.6 RL training
OpenRouter· 09-21
Heat 128
02
Qwen-Image-2.1, a 7B-parameter open-source model for generation, editing, and transparency
shao__meng· 09-21
Heat 128
03
Open source Kev model released by Cognition team
shao__meng· 09-21
Heat 128
04
Anthropic engineers factor RSA-896 in 10 days using 2,048 GPUs
Blognone· 09-20
Heat 127
05
NVIDIA paper fixes tool calls in duplex speech models
elvis· 09-20
Heat 126
06
Kernel-Level Preemption Framework Targets Rogue Autonomous Agent Containment
arXiv cs.AI· 09-25
Heat 126
07
Precision Choices Break Greedy Decoding Determinism in LLMs
arXiv cs.LG· 09-23
Heat 126
08
Xiaomi releases MiMo-V2.6-Pro model
Clement Delangue· 09-21
Heat 125
09
LlamaIndex Grounded Confidence for Document Extraction
Jerry Liu· 09-22
Heat 124
10
Grok 4.7 flight simulator test
eric zakariasson· 09-21
Heat 124

Papers top 10

10 stories
01
Kernel-Level Preemption Framework Targets Rogue Autonomous Agent Containment
arXiv cs.AI· 09-25
Heat 126
02
Precision Choices Break Greedy Decoding Determinism in LLMs
arXiv cs.LG· 09-23
Heat 126
03
DeepSeek发布DSec沙箱基础设施：单日支撑约300万个智能体训练沙箱
arXiv: DeepSeek· 09-22
Heat 123
04
Minimal-Norm ReLU Network Classification Study Yields Exact Global Optimality Results
arXiv cs.LG· 09-24
Heat 121
05
SkillGym Turns Human Agent Skills into Training Environments
arXiv: DeepSeek· 09-24
Heat 121
06
Spectral Theory Explains Grokking via Weight Decay and Feature Learning
arXiv cs.LG· 09-23
Heat 121
07
Softmax Attention Study Identifies Abstention and Noise Filtering as Missing Primitives
arXiv cs.LG· 09-21
Heat 117
08
研究发现：AI 智能体会按推断的财富为富裕用户推荐更贵选项
arXiv cs.AI· 09-22
Heat 115
09
World Modeling in Transformers
arXiv cs.AI· 09-21
Heat 115
10
FeatLens: Feature-Guided Code Graph Retrieval for Repo-Level Code Generation
arXiv: DeepSeek· 09-23
Heat 110

Top 5 by track

US vs ChinaOpen track →
01
Xiaomi's MiMo-V2.6 RL training
02
Qwen-Image-2.1, a 7B-parameter open-source model for generation, editing, and transparency
03
Open source Kev model released by Cognition team
04
Anthropic engineers factor RSA-896 in 10 days using 2,048 GPUs
05
Xiaomi releases MiMo-V2.6-Pro model
China to ASEANOpen track →
01
Qwen-Image-2.1, a 7B-parameter open-source model for generation, editing, and transparency
02
Alibaba Cloud 为马来西亚 Loomi Entertainment 提供 AI 内容安全护栏
03
Alibaba 发布 Qwen Intelligence 手机智能体平台，荣耀 Magic9 成首个采用者
04
Alibaba Token Hub 在云栖大会分享智能体工作流实践
05
阿里云在 Apsara2026 发布智能体云战略升级，围绕 Model、Harness、Context 三大支柱
PolicyOpen track →
01
Kernel-Level Preemption Framework Targets Rogue Autonomous Agent Containment
02
OpenAI 内部模型称解出纳维-斯托克斯方程，27 位菲尔兹奖得主联名质疑
03
Anthropic Explains Why Claude Opus 5.5 Agents Stop Mid-Task
04
Meta Explains Muse Secure VM Cloud Computer Architecture
05
Dutch Government Builds Microsoft Alternative on NixOS via DAWO
ASEAN capitalOpen track →
01
李开复对话东南亚声音：开源与闭源模型之争关乎数据主权
02
MDEC CEO：数据中心投资只是起点，要让中小企业用上算力
BuildersOpen track →
01
Xiaomi's MiMo-V2.6 RL training
02
Qwen-Image-2.1, a 7B-parameter open-source model for generation, editing, and transparency
03
Open source Kev model released by Cognition team
04
NVIDIA paper fixes tool calls in duplex speech models
05
Kernel-Level Preemption Framework Targets Rogue Autonomous Agent Containment
Heat combines model scoring, source weight and multi-source coverage. Charts are not hand-edited. · Read the weekly · See trends