Lydia Hallie ✨@lydiahallieAI 评分6161
holy fuck. so auto compact basically loads last 1M tokens into context by default on claude? and here i thought they were summarising and reducing the context loads. no wonder people used to burn anywhere from 2%-15% 5h limits simply for auto compacting their sessions.
Lydia Hallie ✨@lydiahallieAI 评分6060We just let in every Pro and Max user from the Claude Code Projects waitlist! If you're new to Claude Code Projects, here's a 4 minute walkthrough to get you started:
StepFun@StepFun_aiAI 评分5555@StepFun_ai just announced Step 5 Preview, and it's FREE in Kilo for a week: • 600B parameters, 27B active per token • 1M-token context + vision • highlights: strong coding and finance performance at lower cost. Try it and let us know what you think!
Alexander Embiricos@embiricoAI 评分5252An update for builders using Codex on Windows: We’ve built a new sandbox mode using @Microsoft’s Execution Containers (MXC) for faster setup, stronger network enforcement, and granular file access controls. Requires a compatible Windows 11 device.
ClaudeDevs@ClaudeDevsAI 评分4747我们在一个 11.6 万行的代码库中埋了 70 个 bug。3 次运行中，单个智能体分别找到了 14、15 和 27 个 bug。而一个工作流在 3 次运行中每次都稳定找到 66 个。
Arena.ai@arenaAI 评分5555Arena 发布本周榜单更新：Nano Banana 2.1 在三个 Image Arena 模式进入前六，其中 Multi-Image Edit 排第 4（1431 分）。
Tessl 发布博客拆解智能体代码审查的完整结构，包括标准、视角、发现、裁决、逐条回应与记忆六个环节。其 Tessl Code Review 内置正确性与数据完整性、安全与隐私、规模与韧性、可维护性与代码质量四个视角，以 skill 形式打包并并行运行，标准以 context 形式存放在仓库中。该工具可免费安装到代码仓库，在下一个 pull request 上运行首个循环。
Simon Willison 用 ChatGPT 桌面应用 Codex 标签页的语音模式，在做饭的约半小时内基本靠口头指令为博客开发出 Newsletters 索引功能，使用模型为 GPT-6 Astra High。
JetBrains 发布编程 AI 模型 Mellum2.1，延续 Mellum2 的 12B 混合专家架构、2.5B 活跃参数，仍以 Apache 2.0 许可证开源，重点增强智能体编程能力。
Rohan Paul@rohanpaul_aiAI 评分5353一位独立研究者用 Claude Code（Opus 5.5 和 Fable 5.1）分析 NASA TESS 数据，发现一个自 2018 年起未被报告的行星候选体。
作者用 Claude Max 和 Codex Pro 的多智能体编排，历时约 3 个月、消耗估计 600-700B tokens 将一款热门第一人称射击游戏反编译为 C++，最终 99% 函数重建完成、83% 字节级精确，游戏可完整运行。
Databricks 博客介绍如何用 Lakebase Postgres 的数据库分支能力支持 Agentic SDLC。Lakebase 可在不到一秒内分支任意大小的数据库，采用 copy-on-write 共享父分支数据、支持 scale-to-zero。
Claude Code v2.1.295 为 command 和 HTTP hooks 新增 onFailure: "block"，hook 无法启动、超时或异常退出时将阻断操作而非放行。
Arena.ai@arena精选AI 评分7171隆重推出 Claude Haiku 5.5：我们迄今为止发布的最便宜、最快、最强大的小型模型。 平均而言，它的运行成本比 Claude Haiku 4.5 低约 75%。
推荐理由：榜单方给出 Haiku 5.5 在 Code Arena 的实际得分与各档 Claude 模型的成本分差，便于横向权衡性价比。
Rohan Paul@rohanpaul_aiAI 评分5757Stop pinging me and let my Codex talk to yours. That's why we built @alook_ai — think of it as Grok Bot meets Discord in one open-source product. Take alook at https://alook.ai, or host it yourself. 1/6
Codex 发布 v0.162.0，在启用 worktree 功能后，可从受信任的本地项目创建和列出托管 Git worktree，并支持在智能体 Command Center 中用 p 键置顶任务。该版本还新增 /copy 导航与复制 transcript 区块、Ctrl+Insert 复制选中内容，并可通过 tui.mouse_scroll_speed 调节滚轮速度。
JetBrains 发布面向编码 Agent 的开源思考模型 Mellum2.1（checkpoint 为 Mellum2.1-12B-A2.5B-Thinking），总参数 12B、每 token 激活 2.5B，以 Apache 2.0 协议上架 Hugging Face。
Augment Code 宣布将 Cosmos、Auggie CLI、Code Context Engine 及相关技术出售给 Harness，相关团队将转入 Harness 继续推进这些产品。文中给出的理由是 Harness 的集成平台与 Autonomous SDLC Platform 能更有效地向客户交付这些能力，团队将继续朝软件工厂方向改进大规模 AI 辅助软件开发。
Tessl 发布 Code Review 产品，将审查标准定义为仓库内的 lens 技能，按文件路径路由，团队可在 PR 中查看和修改这些标准。与黑盒审查工具不同，审查沉淀的知识可复用于代码生成与维护；Tessl 自称其 PR 量增加 5-10x 时旧工具账单 80% 来自超额费用，该产品面向智能体高频开发场景。
Tessl 工程博客阐述如何用充足上下文让 Agent 承担需要判断力的软件工程工作，并以 API 迁移为例说明 Agent 如何依据兼容性政策在多种方案间自主选择。
Ant Ling@AntLingAGIAI 评分4646Ling-3.1-flash is now free on OpenCode 560B total · 25B active · 262K context inclusionAI’s latest model
Ant Ling@AntLingAGIAI 评分4242Ling 3.1 Flash from @AntLingAGI is on AI Gateway, free through Oct 13. Built for coding and multi-step analysis. 𝚒𝚗𝚌𝚕𝚞𝚜𝚒𝚘𝚗𝚊𝚒/𝚕𝚒𝚗𝚐-𝟹.𝟷-𝚏𝚕𝚊𝚜𝚑 https://vercel.com/changelog/ling-3-1-flash-is-now-available-on-ai-gateway
Rohan Paul@rohanpaul_aiAI 评分5555微软在旧金山发布会宣布，GitHub Copilot 将于本月底前支持本地 AI 模型推理，开发者可在云端模型与设备端模型之间自动或手动切换，设置覆盖 GitHub Copilot CLI、Copilot 应用与 Visual Studio Code。
Rohan Paul@rohanpaul_aiAI 评分7373
Today marks a new chapter for Windows, as we bring unmetered intelligence to every desk and every home, and make every PC a place where agents can work securely on your behalf. Some highlights of what we announced: • MAI-Code-1.1 Flash: 137B parameter coding model w/ 256K context window, which is now optimized to run on your PC! • GitHub Copilot now hands off work to local models like MAI-Code-1.1 Flash, helping projects cost a lot less without sacrificing quality. • With Hybrid Intelligence, Copilot can now take action directly on the PC and keep sensitive work on your device. • And with Code in Copilot, you can essentially build any software you need on your desktop, without any cloud token spend, and it’s just super at it. You’re no longer limited to what’s in an app store! Your PC becomes an infinite software factory. • Security is foundational to all this, which is why we are also bringing together Windows and Agent 365 so agents can work within secure boundaries on-device, including MXC a local sandbox for agent execution. Windows becomes your secure agent box! • All this comes to life on a new generation of devices, like Surface Laptop Ultra, powered by NVIDIA RTX Spark. Can’t wait to see what you build with all this.
Satya Nadella@satyanadellaAI 评分6363微软在旧金山活动中宣布计划将 130B 参数的 MAI Code 1.1 Flash 模型引入 Windows 11 设备。
Arena.ai@arenaAI 评分5959Mistral Large 4 by @MistralAI just landed in the Code Arena: WebDev at #45. With 1534 pts, this is a +304 pt improvement from Mistral Large 3 at #130! This release also marks a 271-point improvement over its next-best-performing variant, Mistral Medium 3.5, at #122. Mistral Large 4 delivers performance within just 2 points of Claude Opus 4.8 (High) at a nearly 6× lower blended price: $3.47/M vs. $20/M tokens. Nearby but higher-performing models are available for less, keeping it just off the Code Arena: WebDev Pareto frontier. @MistralAI announced its open weights will be released at the end of October. Stay tuned for its score on Arena within open models, and congrats to the team on this release!
Rohan Paul@rohanpaul_aiAI 评分4848Introducing Commenting in Bolt. Click anything in your preview, leave a comment, and send it straight to Bolt to build. No more screenshots, Slack threads, or Loom videos someone has to turn into fixes. Try it now → https://bolt.new
法国公司 Mistral 发布 1 万亿参数开源权重模型 Mistral Large 4，昵称 Le Chonk，目前为预览版，正式版将在月底前推出。该模型针对编码、网络防御以及制造、金融、电气工程等垂直领域优化，Mistral 称其为在中国之外开发的最强开源权重模型，与部分闭源模型非常接近，并称模型为从零训练而非知识蒸馏。
Rohan Paul@rohanpaul_aiAI 评分6161Apple 论文《How Much of a Harness Does a Strong Agent Need for Autonomous ML Engineering?
Roshan Sadanani@roshan_sAI 评分5757Cursor 推出新的远程控制功能，用户可通过 Cursor iOS 应用从手机控制电脑上的 Agent。支持查看进度、回复，或直接启动新任务。
You can now control agents on your computer from your phone. Check in, reply, or start new tasks from the Cursor iOS app.
Codex 0.161.0 将 GPT-6.1 Sol 设为内置目录和 Amazon Bedrock 目录的默认模型，Bedrock 同时支持多智能体 V2 与 Ultra 推理。新增 /mcp login 终端登录 MCP 服务器、语音对话设备选择，以及通过 --enable cli_daybreak 或 features.cli_daybreak=true 启用的 Daybreak 功能。
Boris Cherny@bchernyAI 评分4343Prompt
Boris Cherny@bchernyAI 评分5353I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached. TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt. I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted. Is formal verification the future of coding (or at least, bug finding)?
Claude Code v2.1.292 为 claude plugin install 新增 --marketplace 参数，可在同一策略校验下自动添加市场并安装插件；Agent 工具新增 effort 参数，可指定子智能体的努力度等级。该版本还修复了沙箱命令读取 /ultrareview 暂存文件、UNC 路径绕过文件读取权限提示等多项安全与稳定性问题。
