模型发布/更新
-
SenseNova U1 technical report released, opening model weights based on MoE architecture
A team led by co-founder and chief scientist Dr. Li Mu released the SenseNova U1 technical report, detailing its…
-
IBM releases Granite Embedding Multilingual R2 open-source multilingual embedding models
IBM has released two Apache 2.0 open-source multilingual embedding models: the 97M-parameter compact model…
-
Ant Group's inclusionAI releases trillion-parameter reasoning model Ring-2.6-1T
Ant Group's inclusionAI has released the flagship reasoning model Ring-2.6-1T, with a parameter scale of one trillion,…
-
Kimi K2.6登顶金融智能体基准榜首
Kimi K2.6 现已成为 Finance Agent Benchmark V2 开源权重排名第一。 [引用 @ValsAI]:AI 能胜任金融分析师的工作吗?
产品发布/更新
-
Kimi launches web bridge extension supporting multi-platform interaction
Meet Kimi Web Bridge — Kimi's browser extension. Agents can now interact with websites like humans: search, scroll,…
-
Codex mobile preview launches for remote coding task management
Users can now use Codex anytime, anywhere through the ChatGPT mobile app. The feature supports real-time monitoring,…
-
OpenCode and Qwen 3.6 Plus free again
OpenCode x Qwen 3.6 Plus - free again. Last time you treated our capacity like an all-you-can-eat buffet. We found…
-
Codex introduces automation hooks and programmatic tokens
Codex is becoming easier to automate and customize around users' code. Core updates include 'hooks,' which allow…
-
Luma Agents streamlines end-to-end e-commerce asset generation
More products. More markets. More formats. No more bottlenecks. Define the requirements. Set the style. Luma Agents…
-
SuperGrok Heavy限时六折,Grok Build开放测试
SuperGrok Heavy 现提供约67%的半年折扣,即每月仅需99美元(原价300美元)。 强烈建议升级至Heavy版本,并试用Grok Build的测试版。
-
xAI launches early beta of Grok Build
xAI has launched the early beta of Grok Build for SuperGrok Heavy subscribers. It is a new coding agent and…
行业动态
-
Anthropic and Gates Foundation launch $200M partnership focused on global health and education
Anthropic and the Gates Foundation have established a four-year, $200 million partnership to jointly advance global…
-
百度推进智能体布局,以日活为关键指标
百度推进智能体组合以拥抱智能体时代,主张将日活跃智能体作为关键指标 https://www.prnewswire.com/news-releases/baidu-advances-agent-portfolio-to-embrace-th…
-
AI boom sparks backlash: seven in ten Americans oppose data centers near their homes
A Gallup survey shows that as many as seven in ten Americans oppose building data centers near their homes, a sharp…
-
MiMo V2.5 Pro takes third place on Design Arena
MiMo V2.5 Pro has just taken third place on @DesignArena! 🎉 MiMo V2.5 Pro (Thinking) climbed 8 places over MiMo-V2.5…
论文研究
-
NousResearch推出Token Superposition Training技术,显著加速大语言模型预训练
NousResearch发布了Token Superposition Training(TST),这是一种改进标准大语言模型预训练流程的方法。该技术无需改变模型架构、优化器、分词器或训练数据,即可在相同计算量(FLOPs)下实现2-3倍的…
-
Teaching vision-language models to speak cinematic language
A research team worked with more than 100 professional creators over the course of a year to build a video caption…
技巧与观点
-
Founder's playbook: Building an AI-native startup
Anthropic has released a practical guide for AI-native startups, aiming to reshape the four core stages of the 2026…
-
Cutting token consumption by 61%: Tencent open-sources Agent Memory
Tencent Cloud has open-sourced TencentDB Agent Memory, aimed at solving the problems of context windows filling up…
-
Using Claude Code effectively in large codebases: Best practices and getting started
Claude Code has been successfully deployed in multi-million-line monorepos, legacy systems, and distributed…