关注 AI 研究者、开发者与机构的动态
Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon. https://t.co/GYzoKbxS23
推荐理由：材料说明 Claude Code Projects 的开放范围与云端运行方式，末尾附有对下一版 Opus 的期待。
Noah Zweben@noahzwebenAI 评分5555Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon.
试试看：在 https://t.co/GBG0dyLver 上按 ⌘K 了解更多：https://t.co/GBG0dyLver
Popular 是过去一周按 token 用量计算的最常用模型。Trending 是按 token 用量周环比增长最快的模型。两者都直接来自排行榜数据，所以搜索是进入排行榜的实时快捷方式。
Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon. https://t.co/GYzoKbxS23
推荐理由：Projects 把并行线程调度与云端持续运行合到一处，读者可据此判断编码任务分派方式的变化。
Chinese frontier labs are being priced at radically higher revenue multiples than their US peers, per new research from Rhodium. Moonshot at roughly 167x ARR and DeepSeek at 163x, versus 34x for OpenAI and 21x for Anthropic. Moonshot’s reported valuation alone surged from $4Bat the end of 2025 to $50B by August 2026. These are specifically valuation-to-ARR multiples, not ordinary trailing-revenue multiples. The reason is obvious. The denominator is especially weak in China because open-weight distribution limits direct monetization, model prices are dramatically lower, and Chinese customers historically pay less for SaaS and cloud services
OpenAI and Anthropic earn 10 times more than all Chinese AI models combined, per Rhodium research. OpenAI at roughly $40B in ARR and Anthropic at $65B, while ByteDance’s large-model business is about $4B and Alibaba’s Bailian MaaS about $2.3–$2.4B. And the reason is, models from China are mostly open source. Chinese models recorded more than 2B Hugging Face downloads in 2025, or 41% of the platform total. i.e. users can self-host them or access them through third parties without much revenue flowing back to the original developer.
好观点！测试之后，Jev 感觉像是构建可靠 AI 系统的一个重要原语。我认为还有更多原语有待发现，它们能让基于 LLM 的智能体变得更好、更快。https://t.co/ftZ7LvL7LH
Projects now run from one conversation, starting in Claude Code. You describe what needs doing, and Claude directs parallel threads that keep working after you close your laptop. In beta today for select Pro and Max users in cloud sessions; coming to all Claude users soon. https://t.co/GYzoKbxS23
LimiX-2 一种面向通用结构化数据智能的上下文机制网络 paper: https://t.co/Xf7nmZzLJk https://t.co/dMCpDtK48J
抱歉，您提供的主推文内容只有一个链接（https://t.co/xikdevAj9M），没有实际的推文文字。请提供推文的完整文字内容，我才能为您翻译并取标题。
We’re rolling out effort controls in Computer’s model selector. Effort presets combine the orchestrator model and reasoning depth to control how deeply and efficiently Computer works through a task. Available now on web. Coming to mobile and desktop soon. https://t.co/7Kuj6vavcf
线程在云端运行，即使你的电脑不在线。对本地工作流的支持很快就会到来。 更多内容见文档：https://t.co/NhA6H16fGF
线程在云端运行，所以即使你的电脑离线，它们也能继续工作，但目前还无法访问你电脑或内部网络上的文件或工具。本地支持很快就会推出。
