Latest

Every published item, in order · 6305 items
Nov 11
ModelsUS vs ChinaMulti-source01:37
Chinese contentMustafa Suleyman 宣布 Microsoft AI 发布 MAI-Transform 1 模型

Microsoft 的自研模型团队又有新动作,Suleyman 亲自转发官方页面。链接截断了具体型号,点进去能确认是哪个模型、跑分和定价如何。想看 Microsoft 怎么摆脱对 OpenAI 依赖的人可以关注这条线。

Event
Mustafa Suleyman@mustafasuleymanSource
ModelsUS vs ChinaMulti-source88°01:29
Chinese contentGemini 4 Carbon 上线,面向 Google AI Ultra 订阅者和付费开发者开放

Google 把最强的 Gemini 4 Carbon 直接开放给 Ultra 订阅用户和付费 API 开发者,不用再等内测。如果你在对比 OpenAI 或 Anthropic 的旗舰模型,现在可以拿同一套提示词去 API 里实测。做应用开发和重度用户的都该看看接入门槛和价格。

Event
The Rundown AI@therundownaiSource
ModelsUS vs ChinaMulti-source87°01:26
Chinese contentGoogle 向所有付费开发者开放 Gemini 4 Carbon

Google 把自家的 Carbon 全量放开了,API 价格还是 2/10 美元那档。三个基准跑到第一还比对手便宜一半以上,做 Agent 和自动化任务的开发者可以直接拿它和 Argon 对比跑分,看看多花的推理 token 划不划算。

Event
rohanpaul_ai@rohanpaul_aiSource
Chinese contentGoogle 开放 Gemini 4 Carbon:付费 API 今日可用,Ultra 订阅每月 99.99 美元起

Google 这次把 Carbon 的引入价定到 11 月 30 日,输入 2 美元、输出 10 美元,之后翻倍。跑分上被 Claude Fable 5.2 压过一头,但每个任务花费只有 Fable 的一半,性价比明显。做长任务 agent 或自动化流程的开发者可以趁引入期试试,Antigravity 的 /swarm 多 agent 命令也值得上手。

宝玉@doteySource
ModelsBuildersMulti-source81°01:24
Chinese contentGemini 4 Carbon 上架 Cursor,CursorBench 高努力档得分 60.4%

写代码的朋友可以试试了,Gemini 4 Carbon 已经进了 Cursor,在 Cursor 自家的 CursorBench 上高努力档拿了 60.4% 排第一。之前一直用 Claude 或 GPT 写码的话,正好切过去跑跑自己的项目,看看谁顺手。

Event
Official accountCursor@cursor_aiSource
ModelsBuildersSelected01:23
BEHAVIOR Challenge at CoRL 2026 crowns Huawei Noah EAI Team

Nearly half of long-horizon household tasks now complete successfully, up from 12% a year earlier — a pace of improvement that suggests embodied agents are approaching practical utility faster than most roadmaps assumed. Teams building home or service robots should treat the BEHAVIOR task suite as a reference evaluation and study the winning approaches. The all-Chinese podium also matters for anyone tracking where embodied AI talent and compute are concentrating.

Event
Official accountFei-Fei Li@drfeifeiSource
ModelsUS vs ChinaMulti-source01:19
Chinese contentFable 5.2 发布:芯片设计与生物学能力提升,RSI 任务被有意削弱

GDPx 是少数由领域专家出题的基准,能看清 Fable 5.2 到底强在哪。有意思的是 Anthropic 主动压低了 RSI 任务的表现,分数上直接看得出来。如果你拿模型做芯片、生物或写代码相关工作,这份分项数据很有参考价值。

Event
Dylan Patel (SemiAnalysis)@dylan522pSource
ModelsUS vs ChinaMulti-source01:17
Chinese contentMicrosoft 发布 MAI-Transcribe-2:0.10 美元转录一小时音频

微软把语音转录价格打到 0.10 美元一小时,速度还比 GPT-Transcribe 快 10 倍。做会议记录、播客字幕、客服录音分析的工具团队可以立刻去试试,成本敏感的项目用它能省不少钱。

Event
Mustafa Suleyman@mustafasuleymanSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
ModelsBuildersMulti-source87°01:15
Chinese contentAnthropic 发布 Fable 5.2 与 Mythos 5.2,编码基准超 GPT-6 Astra

Anthropic 这波同底座双版本挺有意思:Fable 面向开发者,Mythos 只给安全机构。FrontierCode 2.0 跑分直接压过 GPT-6 Astra,价格还降了 20%。做硬核编码和自动化工作流的人可以去试试,1M 上下文加多 agent 协调是亮点。

Event
berryxia@berryxiaSource
ModelsBuildersMulti-sourceSelected01:15
MiniMax-H3-Image Gains Day-0 vLLM-Omni Support

Day-0 support on AMD GPUs, not just NVIDIA, is the notable detail, since open image models frequently arrive CUDA-only. Teams building multi-reference editing or high-resolution text-rendering pipelines can self-host without vendor lock-in. Developers comparing against Qwen-Image or closed APIs should benchmark its 16MP native output and 9-image reference limit in their own workflows.

Event
Official accountvLLM@vllm_projectSource
ModelsUS vs ChinaMulti-source01:12
Chinese contentClaude Fable 5.2 以 67% 登顶 VIALS-AA 科学视觉推理基准

Artificial Analysis 拿 Handshake AI 的 VIALS 基准做了生命科学视觉推理实测,Claude Fable 5.2 拿 67% 压过 Gemini 4 Carbon 和 GPT-6 Astra。有意思的是它平均 37 次工具调用对 Astra 的 8 次——做科研图像分析、依赖多步工具链的开发者可以看看这种'多调用'策略是不是更适合自己的场景。

Event
Official accountArtificial Analysis@ArtificialAnlysSource
ModelsUS vs ChinaMulti-source79°01:12
Chinese contentClaude Fable 5.2 登顶 AA-3D 3D 构建基准

Anthropic 新的 Claude Fable 5.2 在 3D 交互构建基准上比自家上代高出 132 分,也压过 GPT-6 Astra 和 Gemini 4 Argon 一大截。做 3D 内容生成、交互原型的开发者可以换它试试,Elo 分差说明优势不是测试波动。

Event
Official accountArtificial Analysis@ArtificialAnlysSource
ModelsBuildersMulti-source01:11
Chinese content用户用一段提示词让 Fable 5.2 一次性生成画面媲美 AAA 的 FPS 游戏Demo

这条提示词原文很短但很讲究:指定 FPS 类型、要求 AAA 级画面、还主动砍掉关卡规模。用 Fable 5.2 一次生成就跑通了。做生成式游戏或想学提示词写法的都该拿去复现试试。

Event
Andrew Curran@AndrewCurran_Source
ModelsBuildersMulti-sourceSelected01:08
Tencent Youtu releases Qwen3.8-27B-DSpine-b16 draft model

Draft-model engineering is quietly becoming a competitive lever: the gain comes from a better-matched draft model rather than the target model itself, so teams running Qwen3 27B can swap this in and roughly double effective throughput with no accuracy trade-off. Inference engineers and startups serving Qwen at scale should benchmark it against their current MTP setup, ideally on Hopper cards where FlashAttention 3 is available. Comparable accelerations typically require paid inference providers; this one is a free HuggingFace download.

Event
karminski-牙医 (AI工具)@karminski3Source
Chinese contentPAN 2 开源世界模型发布,vLLM-Omni 提供 Day-0 推理支持

PAN 2 团队开源了一个按时间、物理、空间来预测场景变化的世界模型,能做文生视频和图生视频。vLLM-Omni 第一天就支持它,想在自己服务器上跑视频生成或做真实世界模拟的开发者可以直接用开源推理栈部署了。

Official accountvLLM@vllm_projectSource
Chinese contentMiniMax 开源图像模型 H3-Image:10 秒生成 16MP 4K 图像

MiniMax 放了个开源的图像模型 H3-Image,10 秒就能出 4K 图,还能塞 16 张参考图做编辑。对比闭源的 Midjourney、Flux,这是开源权重可以直接下载部署的,走 API 还有首周三折起步的五折。做电商图、海报批量生成的开发者可以上手试试。

MiniMax_AI@MiniMax_AISource
ModelsUS vs ChinaMulti-source01:03
Chinese contentFable-5.2 发布,博主对比它与 DeepSeek 的生成成本:40 美元 vs 1.4 元

博主拿刚出的 Fable-5.2 和 deepseek-v4.1-flash-1017 跑了同一个 demo,一个花 40 刀,一个只花 1 块 4,效果对比全在帖子里。做生成类应用、要算成本账的开发者可以直接看图判断值不值,这种带真实账单的横评不常见。

Event
karminski-牙医 (AI工具)@karminski3Source
ModelsUS vs ChinaMulti-source01:02
Chinese contentFable 5.2 在 Remote Labor Index 上完成率达 41.3%,为 5.1 版本的逾两倍

Fable 5.2 在 Remote Labor Index 上把达标率从 17.9% 拉到 41.3%,翻了一倍还多。这个基准用的是真实付费自由职业项目的简报,比刷榜更能反映实际干活能力。做外包开发、设计的同学应该看看,你的活可能已经有人(和模型)在比价了。

Event
kimmonismus@kimmonismusSource
ModelsBuildersSelected01:02
Muse Spark Open Hits 433 tok/s Per User on TPU v7 Megakernel

The notable part is that a 816B MoE model now runs end-to-end as one Pallas program on a single TPU host through SGLang and TorchTPU, not through a closed inference stack. Teams building long-context agent products should study the deep dive for fp8 KV and session-aware cache strategies they can replicate. Compared with CUDA-only megakernel efforts, this demonstrates the PyTorch-to-TPU path has matured enough for production serving.

Official accountPyTorch@PyTorchSource
ModelsUS vs ChinaMulti-source01:00
Chinese contentClaude Fable 5.2 在科学工程领域成绩单:硬件设计、基因疗法、物理难题多点开花

大家都在聊它的写代码能力,其实这份成绩单更值得看:设计电路板一次跑通、34 个病毒蛋白壳建模对 28 个、13 个材料计算从 3.3 小时压到 4 分钟以内。做硬件、生物、材料方向的工程师可以对照自己的领域看看能用上多少。

Event
kimmonismus@kimmonismusSource
ModelsUS vs ChinaMulti-source74°00:58
Chinese contentEthan Mollick 透露 Gemini 4 Carbon 早期体验:可把视频片段复刻为 3D 场景

Ethan Mollick 拿到 Gemini 4 Carbon 早期权限了,他说这代比 Argon 强,而且思路和 Claude、GPT 不一样。最好玩的是丢一段视频进去,它能用 Three.js 复刻出 3D 场景。想跟进 Google 这条线的开发者可以先看看实际体验反馈。

Event
Ethan Mollick@emollickSource
ModelsBuildersSelected00:54
Robbyant releases LingBot-Touch open-source tactile model under Apache 2.0

The non-obvious finding is that masked-latent prediction beats pixel reconstruction precisely because tactile data is sparse and noisy — a lesson transferable to other low-signal sensor modalities beyond robotics. Researchers building touch-equipped manipulators or multimodal perception stacks can adopt the Apache 2.0 weights and the curriculum-masking recipe directly. Compared with MAE and VAE baselines trained on the same data, the JEPA encoder is the one that scales with training hours, which matters for anyone planning long pretraining runs.

rohanpaul_ai@rohanpaul_aiSource
ModelsBuildersMulti-source84°00:53
Chinese contentGoogle DeepMind 用 AI 数学证明 Signal 代码安全性

Google DeepMind 用几十个 Gemini Carbon 智能体写了 650 万行 Lean 证明,4 周干完过去专家团队几年的活,还故意塞回 2 个旧漏洞验证系统真能抓出问题。做安全、写关键系统的工程师都该看看这个验证思路。

Event
rohanpaul_ai@rohanpaul_aiSource
ModelsUS vs ChinaMulti-source00:50
Chinese contentClaude 独立完成一项尘封 23 年的黑洞物理计算

一个挂了 23 年没人算完的物理推导,这次由 Claude 基本自己推完,物理学家核验后确认结论成立。它不是又一个刷 benchmark 的故事,而是 AI 第一次产出经人工确认的基础物理新知识,做科研的人可以直接参考这条工作流。

Event
rohanpaul_ai@rohanpaul_aiSource
ModelsBuildersSelected00:49
Ant Group open-sources LingBot-Touch tactile foundation model for robots

The key shift is that touch becomes hardware-agnostic: tactile representations now transfer across sensors and dexterous hands the way vision backbones do. Robotics teams building manipulation systems no longer need to collect per-hardware tactile datasets, cutting one of the costliest steps in embodied AI development. Developers and labs working on dexterous manipulation or cross-embodiment models can inspect the open-source code and data pipeline directly. For ASEAN robotics teams, an open Chinese-origin foundation model offers a practical alternative to building proprietary tactile stacks.

AIGCLINK@aigclinkSource
ModelsBuildersMulti-source00:48
Chinese contentAnthropic 发布 Claude Fable 5.2,1000+ 子Agent 协作开发出可启动的操作系统

Anthropic 用 Claude Fable 5.2 让 1000 多个子Agent 协作,从零写出能在笔记本上启动的操作系统原型,跑了整整 8 天。比起单模型干活的思路,它展示的是 Harness 编排加多Agent 分工的玩法,做 Agent 框架和自动化开发的工程师很值得看看。

Event
AIGCLINK@aigclinkSource
ModelsUS vs ChinaMulti-source00:45
Chinese contentFable 5.2 发布演示现场证明费马大定理

Anthropic 拿费马大定理证明当 Fable 5.2 的发布演示,这个数学题连人类数学家都要啃好几年。如果你关心前沿模型的数学推理上限,或者想看发布演示还能怎么玩,这条值得点进去。演示者还预告 aie code talk 上有更大的动作。

Event
swyx (AI Engineer)@swyxSource
ModelsBuildersMulti-source72°00:44
Chinese contentGoogle 发布 Gemini Robotics 2.5 物理世界模型

Google 把 Gemini 模型搬进了真实工厂,Apollo 2 机器人给奔驰分拣零件准确率 96%。和 Figure、Tesla 自研路线不同,Google 走模型授权模式。做具身智能和工业自动化的团队可以关注它会不会开放 API。

Event
TestingCatalog@testingcatalogSource

Only the latest 2000 items are listed; for older ones see the daily archive