Latest

Every published item, in order · 29225 items
Nov 10
Chinese contentMeta 发布最小模型 Muse Ember:单价对标 Claude Haiku 5.5,跑分成本仅三分之一

Meta 这次发了个小模型 Muse Ember,专给 agent 打杂用,价格和 Claude Haiku 5.5 持平但每道题花费只有它的三分之一。写代码它比 Haiku 强,改代码合入项目反而弱一点,还有个亮点是几乎不瞎编。给 agent 做主从搭配的开发者值得对比一下各家小模型的实测账单。

宝玉@doteySource
ModelsUS vs ChinaSelected71°03:18
Meta Muse Ember (Max) tops Code Arena WebDev small-model ranking at 1608 pts

The benchmark gap over Haiku 5.5 and GPT-6 Luna is modest in points but wide in cost per completed task, which matters most for teams running high-volume codegen. Developers building website or app scaffolding pipelines can test Ember against Haiku on their own workloads to validate the claimed 2x cost advantage at equal API rates. Volume API buyers should watch whether actual task-level cost, not sticker pricing, becomes the new battleground for small coding models.

lmarena.ai@lmarena_aiSource
Chinese contentAnthropic 发布 Claude Fable 5.2:FrontierCode 得 49.1%,API 降至每百万 8/40 美元

编程评测分数超过了 GPT-6 Astra 和 Gemini 4 Carbon,价格还降了两成,写代码的同学可以直接在 Claude Code 里手动切换试试。Pro 用户 16 号前用 Fable 不额外扣钱,别忘了去设置里领那次额度重置。

宝玉@doteySource
Chinese contentAnthropic 工程师用 Claude Fable 5.2 协调 1000+ 子智能体写出可启动操作系统原型

Anthropic 工程师拿 Claude Fable 5.2 干了件硬核事:1000 多个子智能体连跑 8 天,写出一个能开机、有内核和窗口系统的 OS 原型。做智能体编排的人应该看看这个案例,比一般 coding demo 能说明规模上限在哪。不过消息来自推特,先当参考。

kimmonismus@kimmonismusSource
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
PapersUS vs ChinaSelected03:08
Epoch AI launches Inference Performance Index for chip comparison

The metric shifts chip evaluation from peak compute to real serving capacity, which changes how buyers should read vendor spec sheets. Infrastructure planners, cloud capacity teams and analysts sizing inference fleets can use IPI scores to model how many concurrent users or agents a given GPU deployment actually supports. Compared with the H100, the GB300's 10× inferred advantage shows why last-generation hardware may be far more constrained for agent workloads than raw FLOP/s suggests.

Official accountEpoch AI@EpochAIResearchSource
Chinese contentAWS Bedrock 价格表出现 Claude Fable 5.2 与 Mythos 5.2,定价较 5.1 版下降 20%

在 Bedrock 上用 Claude 的团队可以直接看这条:5.2 版输入输出全线降价 20%,缓存读取只要 $0.20/百万 token。如果你的账单大头在输出 token 或缓存调用,按这个新费率重算一下,一年省下的钱可能不小。

kimmonismus@kimmonismusSource
Chinese contentAirCaps 发布语音转写模型 A5S Async v1 并开源评测基准

AirCaps 发了个会标注「这句话是谁说的」的语音转写模型,VoiceArena 排第一,比 ElevenLabs、AssemblyAI 这些老牌选手错误率还低 41%。难得的是评测代码和数据配方全开源,能自己复现。做会议纪要类产品或者做语音评测的可以盯一下。

elvis@omarsar0Source
Chinese contentArchgen 用 AI 设计 3D 芯片,热仿真速度快 5000 倍

YC F26 的 Archgen 在做 AI 驱动的 3D 芯片设计工具,热仿真比传统 EDA 快 5000 倍,布局引擎还拿了 HRT 挑战赛第一。芯片垂直堆叠后散热和时序是硬骨头,他们让 AI 智能体自己找 bug、修复、重跑。做芯片设计或关注 EDA 格局的人可以看看。

elvis@omarsar0Source
ProductsBuildersSelected02:31
Google expands 8th-generation TPU lineup with TPU 8i Max and 8tx

The real signal here is cadence: four systems in one generation within twelve months signals a shift from fixed silicon cycles to model-driven hardware iteration, compressing the usual ASIC refresh timeline. Teams evaluating inference infrastructure should compare TPU 8i Max against GPU-based serving for long-context workloads, and TPU 8tx against confidential-computing offerings from cloud rivals. PyTorch and Hugging Face developers can now port existing models to TPUs without rewrites via TorchTPU, making a pilot deployment practical rather than a major engineering project.

Sundar Pichai@sundarpichaiSource
ModelsUS vs ChinaSelected02:27
Muse Ember Small Model Hits Efficiency Frontier at One-Third of Haiku Cost

The interesting shift here is that small-model competition is now argued in cost-per-task terms rather than benchmark leaderboards, a framing that maps directly onto procurement decisions. Teams running high-volume workloads like classification, extraction, and batch processing should re-run their cost baselines against Haiku before renewing API commitments. The luna comparison also hints at a tiered small-model lineup that deserves watching as pricing details surface.

Official accountAlexandr Wang (Meta MSL)@alexandr_wangSource
ModelsBuildersSelected02:26
Muse Ember Debuts at 17x Lower Cost Than Muse Spark 1.3

A 17x price cut for near-flagship intelligence reshapes the cost math for token-heavy workloads like coding agents and batch processing. Teams building on Muse models can swap Muse Ember into API calls or Muse Code to cut spend without a full quality drop. Buyers comparing against other small-model tiers now have a new price-performance benchmark to weigh.

Official accountAlexandr Wang (Meta MSL)@alexandr_wangSource

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