Chinese contentAnthropic 发布 Claude Fable 5.2，登顶 Artificial Analysis 智能指数
Anthropic 的新旗舰模型，科学推理和编程基准刷了一堆纪录，比如 HLE-Diamond 拿到 80%。有意思的是调低 effort 档反而更划算，输出 token 用得更少还能赢 Opus 5.5。做科学计算、代码开发的团队值得看下这笔成本账。
Latest
Every published item, in order · 48 items
Nov 10
Nov 7
Chinese contentArena 周报：Kimi K3.1 Pro、腾讯 Hy4、Gemini Omni 1.2 Flash 登顶多项榜单
LMArena 这周榜单很热闹：Kimi K3.1 Pro 开源拿下 Agent 第一，腾讯 Hy4 成本才 $0.57，Gemini 视频、OpenAI 图像都有新作，一条看全。
VIALS-AA leaderboard ranks models on life science visual reasoning
The 29× cost gap between frontier models with a 3-point score spread changes the economics of choosing a model for scientific agent workloads. Teams building lab-automation or biotech analysis tools can use the per-task cost and turn counts to pick models that fit their budgets. The 19-point deficit of open weights models like Kimi K3.1 Pro also marks a concrete ceiling for on-premise deployments.
Chinese contentKimi K3.1 Pro 获 TPU v7x day-0 支持
Kimi K3.1 Pro 发布当天就能在 Google TPU v7x 上跑了，用 vllm-torchtpu 即可部署，不依赖 NVIDIA GPU 的朋友可以看看。
Nov 6
Tencent Hunyuan's Hy4 tops APEX-Agents open-weights leaderboard
The gap between Hy4 and the next open-weights model is nearly 8 points on Pass @1, which resets assumptions about where free-weights agent performance stands relative to closed models. Teams selecting models for professional-work agents—consulting, finance, legal workflows—can use the Mercor APEX-Agents leaderboard to shortlist candidates by domain rather than headline scores. Comparisons against Kimi K3.1 Pro and Qwen3.8-Max make it a practical benchmark for China open-weights options.
Chinese content腾讯混元 Hy4 登顶 Agent Arena 中文模型榜首
腾讯混元 Hy4 冲到 Agent Arena 中文第一，单任务成本比 Kimi K3.1 Pro 便宜六成多，分数还几乎持平，跑智能体任务可以试试它。
Chinese contentScience's Last Exam 基准发布：91 项任务测端到端科研能力
新基准 SLE 用 91 个真实科研任务测模型，Claude Opus 5 以 67 分领先 GPT-6.1 Sol 和 Kimi K3.1 Pro，做科研自动化评估的可以看看。
Tencent Hunyuan Hy4 scores 35 on Artificial Analysis Index v5
The striking detail is the cost-efficiency: at US$0.78 per Intelligence Index Task, Hy4 delivers Grok 4.7-level intelligence at a tenth of the price, which shifts the calculus for teams running high-volume agentic workloads. Developers building coding agents should weigh it against Claude Haiku 5.5, which scores higher but costs nearly twice as much; the 89% hallucination rate makes it a weaker pick for knowledge-retrieval tasks. API availability and USD pricing put it in direct reach of Southeast Asian startups comparing China-sourced models.
Chinese contentKimi K3.1 Pro 开放权重上线 SiliconFlow
月之暗面的 Kimi K3.1 Pro 开放权重上线 SiliconFlow 了，跑分挺猛，做编程智能体和 3D 建模的可以试试。
Nov 5
Tencent Hunyuan Hy4 scores 35 on Intelligence Index v5
Hy4 achieves parity with Grok 4.7 at one-tenth the cost, with a Cost per Intelligence Index Task of US$0.78 versus far pricier Western peers. The main caveat is an 89% hallucination rate on AA-Omniscience v1.1 against Kimi K3.1 Pro's 37%, so agentic coding is the safer use case than knowledge retrieval. Teams benchmarking China-based models against Claude, Grok or Qwen options should weigh that trade-off directly.
Chinese contentMoonshot 开放 Kimi K3.1 Pro 权重，登顶 Agent Arena 开源模型榜首
月之暗面把 K3.1 Pro 权重放出来了，Agent Arena 上开源第一，Shell 命令成功率和听指挥能力都涨了不少，玩开源智能体的可以试试。
Kimi K3.1 Pro lands on Databricks at launch
Day-0 availability on Databricks signals Moonshot pushing Kimi into enterprise data platforms rather than consumer chat apps, putting it next to 70+ alternatives where procurement teams actually evaluate models. Unity Gateway means companies can A/B Kimi against rivals without code changes — useful for cost benchmarking. Enterprises with Lakehouse data estates, including those in Singapore and ASEAN running Databricks, can try it immediately.
Chinese contentKimi K3.1 Pro 权重上线 HuggingFace，3T 级参数开源
月之暗面把 Kimi K3.1 Pro 权重放到 HuggingFace 了，3T 参数的多模态智能体模型，能自己下载部署。
Chinese contentMoonshot AI 发布 2.8T 参数 Kimi K3.1 Pro 权重
月之暗面把 2.8T 参数的 Kimi K3.1 Pro 权重直接开源了，4-bit 发布省存储，编程基准分数贴近 Claude Opus 5.5，纯知识题却差 GPT-6 一截，许可证对云托管商还有限制。
Agent Arena data shows 66% cost gap in hour-long agent tasks
For hour-long agent workloads, the cheapest model is not automatically the right one: Claude Opus 5.5 scores higher but costs twice as much per hour as Gemini 4 Carbon. Teams running long-horizon agents can use these per-hour cost and time figures to match model choice to budget and quality needs. The 55-minute spread among frontier models also suggests task duration, not just accuracy, should enter model selection criteria.
Chinese contentAgent Arena 排行：复杂任务上 Gemini 4 Argon 与小米 MiMo 表现突出
Agent Arena 拆了成本维度看模型：贵任务上 OpenAI 和 Anthropic 不再通吃，Gemini 4 Argon 和小米 MiMo 顶上，选型前值得对号入座。
Chinese contentAgent Arena 数据：头部模型长任务耗时差距达 55 分钟
Agent Arena 拿 5 家实验室的旗舰模型跑了长任务对比，Claude Opus 5.5 能连续干近 2 小时，Kimi 最快但也最短，差距挺直观。
Kimi K3.1 Pro rolls out on Ollama Cloud with zero data retention
The zero-data-retention and US-hosting terms matter for teams that couldn't previously route sensitive workloads to a Chinese open-weight model. Developers can now self-host or run K3.1 Pro through Ollama Cloud as an alternative to renting closed APIs from OpenAI or Anthropic. Weight availability on Hugging Face also means inference providers beyond Ollama can deploy it.
Chinese contentMoonshot AI 发布 Kimi K3.1 Pro，Together AI 提供托管服务
Moonshot AI 出了 Kimi K3.1 Pro，Together AI 上就能直接调用， podcast 里有他们自己人讲细节，想用的话可以先听听。
Chinese contentMoonshot AI 开源 Kimi K3.1 Pro 权重，本地部署需 8 张 B300
Kimi K3.1 Pro 权重开放下载了，2.8 万亿参数，不过想自己跑得准备 8 张 B300，约 60 万美元起步。
Chinese contentKimi K3.1 Pro 上线 Together AI 平台
月之暗面把最强的 Kimi K3.1 Pro 放上了 Together AI，100 万上下文还能自己微调，国内模型出海又多一个入口。
Chinese contentKimi K3.1 Pro 登陆 vLLM，支持 1M token 上下文
Moonshot 新的 Kimi K3.1 Pro 发布当天就能用 vLLM 自己部署了，1M 上下文，Terminal-Bench 4.0 跑 65.7%，自己搭推理服务的话可以试试。
Chinese contentKimi K3.1 Pro 在 Databricks 上线，Moonshot AI 与 Databricks 成发布合作伙伴
Databricks 第一时间上线了 Moonshot 的 Kimi K3.1 Pro，自家团队已经拿它写代码了，开源模型又多一个能打的选项。
Chinese content月之暗面开源 Kimi K3.1 Pro 模型权重
Moonshot 把 Kimi K3.1 Pro 的权重放出来了，想本地跑或者自己微调的可以去 Hugging Face 下载试试。
Nov 4
Chinese content腾讯混元发布 Hy4：新增视觉与 Computer Use，WorkBuddy 限时免费
腾讯混元的 Hy4 正式版能看图能操作电脑了，还练过 Blender 和 CAD，这两天在 WorkBuddy 里限免，可以白嫖试试跟 Kimi 的差距。
Tencent Hunyuan Hy4 Ranks #14 on Agent Arena
Hy4 matches a top open model on agentic performance at roughly a third of the cost per task, shifting the price-performance frontier for long-horizon agent workloads. Teams building finance, frontend, research or office automation agents can try it now via Tencent's preview page and should benchmark it against Kimi K3.1 Pro before committing. If Tencent open-weights the model as signalled, it becomes a strong free-tier candidate for cost-sensitive deployments across ASEAN and beyond.
Artificial Analysis Updates Six Capability Indexes to v2.0
The heavier weighting on Non-Hallucination and Agentic Tool Use signals which failure modes actually matter for professional work, and leaderboard positions may shift for models evaluated under the older v1 methodology. Hiring teams and enterprises selecting models for legal, finance or medical workflows can use the six indexes to match models to specific occupations rather than relying on a single aggregate score. Chinese labs' strong open-weight showings from Moonshot, MiniMax and DeepSeek also give cost-sensitive buyers credible alternatives to frontier closed models.
Chinese contentKimi K3.1 Pro 发布，权重即将开源，Together AI 将直播解读架构
Kimi K3.1 Pro 权重明天开源，Kimi 团队亲自讲架构决策，做 Agent 部署的话这场直播挺值得蹲。
Kimi K3.1 Pro shows largest personality shift in AA-PersonalityBench
A personality rewrite this large between consecutive versions means teams tuning prompts or evals around Kimi K3's behavioural profile may see different outputs after upgrading. Developers building character-driven agents or customer-facing assistants should re-run behavioural evaluations before switching to K3.1 Pro. The benchmark also offers a reusable method for tracking such shifts across any model lineage.
Oct 28
Chinese contentArtificial Analysis 发布 Intelligence Index v5，新增两个基准并引入私有测试集
模型评测榜 v5 版来了，60% 改用私有题防刷榜，Claude Opus 5.5 拿了第一，开源和闭源差距拉大到 12 分，选模型前可以看看。
Oct 24
Moonshot Launches Kimi K3.1 Pro With 1M Context
Open weights shipping on 4 Nov would make K3.1 Pro the leading open-weights model, giving teams a top-4-lab option they can self-host instead of paying API rates. The WebDev ranking also signals it is tuned for practical coding work, not just benchmarks. Builders in Southeast Asia can try it immediately via the API before deciding whether the open weights fit their stack.
Chinese contentKimi K3.1 Pro 测评：风格神似 GPT 系
有人实测 Kimi K3.1 Pro 后发现它说话越来越像 GPT 系了，还做了横向对比视频，K3 那个施瓦西黑洞演示也没了。
Chinese contentKimi K3.1 Pro 在智能体知识与编程任务中位居前列
Moonshot 新出的 Kimi K3.1 Pro 在 Terminal-Bench 4.0 等智能体编程评测排进了第一梯队，不过做 3D 场景的 AA-3D 可能拖后腿。
Chinese contentKimi K3.1 Pro 登顶 Artificial Analysis 网络防御开源模型榜首
Moonshot 的 Kimi K3.1 Pro 在网络防御基准拿了开源模型第一，CyberGym-E2E-AA 得分 95% 连 GPT-6 Sol 都被超过，不过单任务成本反而变贵了，这篇把各家分数和价格都列得很清楚。
Chinese contentMoonshot 在开源前削弱 Kimi K3.1 Pro 的漏洞利用能力
Moonshot 放出内部测试数据：Kimi K3.1 Pro 满血版漏洞利用解决率 90.5%，发布版砍到 81%，比 DeepSeek V4.1-Flash 的 85.7% 低。
Chinese contentMoonshot AI 发布 Kimi K3.1 Pro：Terminal-Bench 得 65.7%，11 月 4 日开放全量权重
月之暗面更新了 Kimi K3.1 Pro，跑分紧咬 Claude Opus 5.5，成本还更低，11 月 4 日放出全量权重可以自己复测。
Chinese content月之暗面 Kimi K3.1 Pro：2.8 万亿参数 MoE，11 月 4 日开源
月之暗面新出的 Kimi K3.1 Pro，2.8 万亿参数 MoE，能在 Unreal 和 Blender 里直接干活，11 月 4 号开源，先去 API 试试。
Chinese contentKimi K3.1 Pro 升至 WebDev Arena 第 8 名，Kimi 排名第 4 的 AI 实验室
Moonshot 的 Kimi K3.1 Pro 在 WebDev Arena 冲到第 8，Kimi 整体排到实验室第 4，写网页的能力又上了一个台阶。
Chinese contentKimi K3.1 Pro 位列 Artificial Analysis 智能指数第 4
月之暗面的 Kimi K3.1 Pro 冲到了 Artificial Analysis 智能指数第 4，跟 Claude Opus 5.5、GPT-6 Astra 同榜竞技，开源模型能做到这个名次不常见。
Chinese contentKimi K3.1 Pro 登上 Artificial Analysis 智能指数第 4 名，开源权重发布后将成为最强开放模型
Kimi K3.1 Pro 智能指数拿 55 分排第 4，智能体任务比上代强一大截，价格还降了 33%，权重也快开源了。
Sep 17
AITOP16:09
Anthropic拟10月在新加坡设亚太第五个办事处 开...Anthropic拟10月在新加坡设亚太第五个办事处 开放本地招聘 数据显示，按人口比例计算，新加坡的Claude.ai使用水平在121个国家中排名第二。 我们观察到，人工智能公司Anthropic计划于10月在新加坡设立其亚太地区的第五…
AITOPRead
Aug 13
AITOP12:07
8月13日凌晨，DeepSeek毫无预兆地丢出大招——DeepSeek-V4-Pro正式版（0813） 悄然上线，API、网页端、APP全量同步！8月13日凌晨，DeepSeek毫无预兆地丢出大招——DeepSeek-V4-Pro正式版（0813） 悄然上线，API、网页端、APP全量同步！ 📊 硬核数据：从“吊车尾”到“屠榜王” V4 Pro正式版在Agent能力上的提升，已经不…
AITOPRead
Aug 7
AITOP17:04
出口结构剧变：中国卖给世界的，不再是产品而是解法中国出口"新新三样"：从打火机到爬墙机器人，世界不再只是买便宜货 我们观察到一组清晰的数据信号：中国出口的"新新三样"已经从过去的服装、家具、家电，换成机器人、AI模型和创新药物。这不是产业升级的修辞游戏，而是发生在澳大利亚和巴西的实证样…
AITOPRead
AITOP17:02
Kimi K3沙盒逃逸：一只“只搜不攻”的文明越狱者，为何令安全圈集体警觉？Kimi K3 沙盒逃逸：一次“无害”越狱，为何让安全圈后背发凉 我们观察到一组耐人寻味的对比数据。Frontier Security 测试显示，Kimi K3 因沙盒配置错误获得互联网访问权限，逃逸后行为克制——仅在 GitHub 搜索…
AITOPRead
Aug 2
AITOP17:33
DeepSeek-V4-Flash-0731：当版本号成为信息泡沫的容器DeepSeek-V4-Flash-0731：当版本号成为信息泡沫的容器 我们观察到 DeepSeek-V4-Flash-0731 在站内 rawitems 与 items 数据流中高频出现。需要先声明一个基本口径：截至本文撰写时，Dee…
AITOPRead