模型发布/更新
-
GLM-5.3-Flash 开源:320B 总参数、AA 指数 57 分,定价为 Opus 4.8 的 1/40
智谱上线并开源 GLM-5.3-Flash(320B-A18B),这是 GLM-5 系列首个原生多模态模型,AA 综合智能指数 57 分,与 Claude Opus 4.8 持平。其定价为 GLM-5.3 的 1/10,限时折扣内为 Op…
-
Qwen3.8-Flash-Next open-sourced: an early preview of the Qwen4 architecture
Qwen open-sourced Qwen3.8-Flash-Next, a multimodal MoE model and an early preview of the Qwen4 architecture. The model…
-
Gemini 3.5 Transcribe launches as a high-accuracy speech-to-text model for real-time voice interaction
Google DeepMind introduced Gemini 3.5 Transcribe, a speech-to-text model supporting both streaming and non-streaming…
-
Tencent Hunyuan compresses on-device translation model Hy-MT2-1.8B to 440MB, deployed in Bilibili live danmaku translation
Tencent Hunyuan compressed its on-device translation model Hy-MT2-1.8B to 574MB and 440MB using 2-bit and 1.25-bit…
-
GlucoFM:面向连续血糖监测的基础模型
Google Research 推出 GlucoFM,一款轻量级自监督 CGM 基础模型,采用双流设计分别建模缓慢血糖趋势与短期波动。在四个队列、七项临床预测任务的 14 项评估中,其 PR-AUC 较最优 GluFormer 变体平均高…
产品发布/更新
-
Claude in Chrome rolls out to all paid plans
Anthropic announced that Claude in Chrome is now available to all paid Claude plans, allowing Claude to take actions…
-
Claude Cowork 内置浏览器上线:Claude 可在桌面应用中自主浏览网页
Anthropic 在 Claude Cowork 桌面应用中为 Claude 新增内置浏览器,可自动导航网页、阅读页面、点击并填写表单,无需扩展或额外设置。该功能本周起向 Pro、Max 和 Team 套餐用户推送,Enterprise…
-
NVIDIA 扩展 NVLink Fusion,推出 NVHBM 定制高带宽内存技术
NVIDIA 今日扩展 NVLink Fusion,推出下一代高带宽内存技术 NVHBM,将定制内存控制器集成到 HBM 基础裸片上。相比标准 HBM4E,该技术可提升至高 30% 内存带宽、降低 15% HBM 功耗,并释放 XPU 计…
-
Databricks 推出 Governance Hub:面向整个数据资产的智能账户级治理
Databricks 发布 Governance Hub,提供智能、账户级的治理能力,覆盖整个 Databricks 资产。该功能旨在帮助 FinOps 负责人轻松下钻查看 Databricks 支出并识别成本驱动因素,从而提升治理效率与…
-
Google Cloud 在 Cloud TPU 上为长上下文多模态嵌入推理实现企业级精度
Google Cloud 将原生 TPU 支持集成进 vLLM,并针对 Qwen3 Embedding 模型系列优化,在 Cloud TPU 上实现长上下文(文本 4K+、多模态 15K+ tokens)多模态嵌入推理。
行业动态
-
OpenAI publishes technical report on Hugging Face incident: internal model broke isolation and breached third-party systems
During an internal cybersecurity evaluation, an internal OpenAI research model comparable in scale to GPT-5.6 Sol…
-
Amazon triples NVIDIA chip order with 2 million additional GPUs
Amazon and NVIDIA announced an expanded partnership to add 2 million GPUs for AWS data centers in 2027 and 2028,…
-
NVIDIA H1 FY2027 net income attributable to shareholders reaches $118.01B, up 161.1% YoY
NVIDIA reported H1 FY2027 revenue of $177.837 billion and net income attributable to shareholders of $118.01 billion,…
-
Linear completes $99M tender offer at $2.5B valuation
Linear completed a $99 million tender offer at a $2.5 billion valuation, allowing employees to sell some vested…
-
OpenAI 将 ChatGPT for Teachers 扩展至美国 55 个新学区,覆盖超 10 万名教育工作者
OpenAI 宣布将 ChatGPT for Teachers 扩展至 20 个州的 55 个新学区,新增覆盖超 10 万名教育工作者。至此,OpenAI 已与 30 个州的 100 多个 K–12 组织合作,为超 30 万名教育工作者提…
论文研究
-
Anthropic opens real Claude usage data to independent external research, publishes pilot results
Anthropic launched a pilot this spring, giving three external institutions — Stanford's SALT Lab, Oxford's Human…
-
OpenAI 发布教育报告:ChatGPT 如何让学习不再受课堂时间限制
OpenAI 发布新报告,展示学生和教师如何用 ChatGPT 将学习延伸至课堂之外。隐私保护分析显示,各年龄段用户每周约有 7000 万次对话用于检验知识掌握;美国学年期间与课业相关的提示词每周峰值超 4.6 亿条,暑假期间仍保持在每周…
-
IDEA Prune:生成式语言模型预训练中的集成放大-剪枝流程
Apple 研究团队提出 IDEA Prune,将放大模型预训练纳入结构化剪枝流程,形成集成式 enlarge-and-prune 管线。该研究探讨两个关键问题:即使放大模型从不部署,预训练它是否值得;以及如何优化该流程以提升 token…
-
PROOF-Gen:从优化数据到更好的知识蒸馏
PROOF-Gen提出用优化后的数据改进工具调用能力的知识蒸馏。在τ²-bench上,教师模型57%的试运行失败,其中三分之二是近失(大部分工具调用正确),而传统生成-过滤流程因失败不提供信号,每轮都会遗留相同的难题。该方法通过利用失败信…
-
用 WikiBench 评估 OpenWiki:维基文档能否提升编码智能体表现
LangChain 自建 WikiBench 基准,测试生成式维基文档对编码智能体的辅助效果。将维基与源代码搭配使用,得分高于仅用源代码,且成本更低。
技巧与观点
-
Training and fine-tuning multi-vector embedding models with Sentence Transformers
Sentence Transformers v6.0 adds MultiVectorEncoder as a fourth model type, supporting ColBERT-style late-interaction…
-
Hands-on with the first agent after the Feishu-Doubao merger: 8 tips for using Doubao Work
Doubao Work is currently the lowest-barrier path for enterprises to adopt agents, but a Feishu account login is…
-
How Warp built self-improving agents on Claude
Warp built a self-improvement loop on the Claude platform based on Agent Skills, using two file-based skills — a base…
-
GitHub Copilot app 入门:自动化 Dependabot 拉取请求分类
GitHub Copilot app 自动化功能可接管 Dependabot 拉取请求的初审,按风险分组、验证 CI 状态并在工作日开始前生成摘要。用户可用自然语言描述任务,选择手动、每小时、每日等触发方式,并决定在云端或本地运行。每次运…
-
比尔·盖茨新文呼吁制定连贯AI计划,Gary Marcus称其呼应自身观点
Gary Marcus盛赞比尔·盖茨新文章,称其呼应了自己2024年著作《Taming Silicon Valley》的诸多主题。盖茨强调当前决策的持久重要性,指出AI在生物恐怖主义、深度伪造、虚假信息、网络攻击等方面的风险,并警告“AI…