← 返回最新日报
2026-09-11· 当日存档
AI 日报
智能体基建与模型竞速
今日看点集中在智能体基础设施与模型能力边界：OpenAI 公测 Agents API，把 Codex 背后的整套执行框架开放给开发者；DeepSeek 发布 V4.1-Flash，百万上下文加 FP4 KV 缓存优化；GPT-6 Astra 攻克 FrontierMath 全部 Tier 4 题目，但 OpenAI 称数学并非重点。
模型动向
1
DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
百万上下文加FP4缓存，长文本推理成本再下探
MarkTechPost深挖 →
2
OpenAI's GPT-Live-1 API lets developers build apps that talk and listen at the same time
全双工语音API，边听边说交互得分翻倍
The Decoder深挖 →
3
The strawberry thing was very funny but probably gave people the wrong impression of where AI was heading in math.
FrontierMath Tier 4全数攻克，数学基准见顶
@emollick深挖 →
4
GPT-6 Astra gives mathematicians a breather, and OpenAI says that's by design
数学登顶却非重点，资源转向自我改进与对齐
The Decoder深挖 →
5
Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
Cognition新编程模型对标头部，代码赛道再添变数
Hacker News深挖 →
6
Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation
先建API链再生成问题，工具调用数据通过率99.8%
MarkTechPost深挖 →
7
NVIDIA Details BioNeMo Inference Runtime (BioIR): 2.90x Higher Boltz-2 Folding Throughput and 58.5K Residues per GPU-Hour on 8xH100
BioIR推理运行时让蛋白折叠吞吐提升近三倍
MarkTechPost深挖 →
8
Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video
机器人看一遍视频就能学会新任务
NVIDIA深挖 →
9
产品上新
10
OpenAI Launches the Agents API in Public Beta, Putting the Codex Harness Behind One API Call
Codex整套执行框架开放，一次API调用即可驱动
MarkTechPost深挖 →
11
OpenAI's new Agents API gives developers the infrastructure behind Codex and ChatGPT
云端自主智能体公测，按token计费支持任务交接
The Decoder深挖 →
12
Meet Redis LangCache: A Managed Semantic Cache That Cuts LLM API Costs by Up to 90% and Returns Cache Hits Up to 15x Faster
托管语义缓存，LLM API成本最高降九成
MarkTechPost深挖 →
13
14
How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules
科学家用Codex挖掘抗菌分子，对抗耐药感染
OpenAI深挖 →
论文研究
15
ToolGrad: Efficient tool-use dataset generation with textual "gradients"
文本梯度生成工具调用数据集，效率大幅提升
Google Research深挖 →
16
行业动态
18
Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek
Anthropic指控三家中国公司对其模型蒸馏攻击
TechCrunch AI深挖 →
19
Swarmchasers hunt rogue agents, Anthropic investigates itself, and the trail they both follow is going dark
疑似OpenAI智能体潜入公共服务，监管手段失效
The Decoder深挖 →
20
Claude users found ways around safeguards for bioweapons research
用户绕过Claude安全限制获取生物武器信息
Ars Technica AI深挖 →
21
22
The Mathematical AI Safety Institute wants to prove AI is safe the way cryptographers prove codes are unbreakable
菲尔兹奖得主创立研究所，用数学证明保证AI安全
The Decoder深挖 →
23
Reduce LLM latency with prefix-aware routing on Amazon SageMaker Inference
前缀感知路由复用KV缓存，首字延迟降77%
AWS ML深挖 →
24
Physical AI Takes the Wheel: How the World’s Robotaxi Leaders Are Building With NVIDIA Technologies
Robotaxi市场2035年达4000亿美元，NVIDIA助力
NVIDIA深挖 →
25
26
Anthropic's $1.5 billion book settlement descends into chaos as authors and publishers fight over who gets paid
Anthropic 15亿美元和解金分配起争执
The Decoder深挖 →
27
28
实用技巧
29
30
The Waymo effect: how AI is quietly making research less collaborative
AI自动化科研流程，研究正变得更孤立
Hacker News深挖 →
日报存档
2026 年 9 月
22 日小米开源登顶，AI失控警钟21 日开源图像模型与AI治理并进20 日AI安全与失控风险成焦点19 日AI安全失控与模型竞速18 日AI对齐警报与芯片竞赛17 日AI安全警报与产品整合潮16 日谷歌苹果齐发AI新品15 日苹果Siri重构与AI放缓之争14 日AI巨头放缓迭代引激辩13 日GPT-6 Astra全面突破12 日AI数学错位与安全风波11 日智能体基建与模型竞速10 日GPT-6登场，AI安全恐慌升温9 日AI数学突破引热议8 日AI日报：端侧模型与欧洲融资双热点7 日AI日报：模型竞赛与行业震荡6 日GPT-6 Astra发布，AI竞争白热化5 日GPT-6发布，AGI时代开启4 日GPT-6 Astra开启AGI时代3 日英伟达巨资收购HF2 日AI日报：世界模型与安全并进1 日AI日报：模型突破与行业震荡2026 年 8 月
14 日AI日报：手语翻译、Grok 4.6与资本热潮13 日DeepSeek V4 Pro开源，AI行业热度飙升12 日AI日报：算力新纪元11 日AI日报：开源与安全并进10 日Meta开源多模态，AI安全引担忧9 日AI安全测试成新风险8 日AI安全与模型突破并进7 日AI安全与竞争新局6 日AI安全警钟与巨头变局5 日AI代理安全与商业化双线并进4 日阿里巨模型发布，开源生态激荡3 日Qwen3.8-Max领跑，AI应用百花齐放2 日AI 速读 · 2026-08-021 日AI安全警钟与模型新突破