Introducing SWE-1.5, our fast agent model. It achieves near-SOTA coding performance while setting a new standard for speed. Now available in Windsurf.

The Cognition and Windsurf team just wrapped up an exceptional Q3. Beyond another quarter of massive revenue growth, what stands out most is our engineering momentum - with launches like Codemaps, Fast Context, and our new SWE 1.5 model Windsurf is also now the first agentic coding platform to be the standard across the engineering organization at one of the world’s largest banks. Every engineer and technologist at the bank will use our AI IDE to accelerate delivery, improve quality, and fuel innovation. Across the Fortune 500, companies are welcoming Devin as a digital employee, assigning it hundreds of thousands of work hours to eliminate technical debt and transforming how engineering teams work. Our mission is clear: own the future of AI-powered software development.

谈谈为什么我更钟情于 @windsurf 而非 Cursor ，以及我所理解的Windsurf的产品哲学 作为 Windsurf 的早期用户，我从它刚推出时就开始使用，一路见证了它的发展。虽然Windsurf的声量不如Cursor和Claude Code，但我想谈谈为什么我依然坚持使用 Windsurf：简单来说，作为零代码基础起步的vibe coder， 让我以较小的精力获得对代码库的掌控感，让vibe coding在可控范围内。 Cursor也是很棒的产品，甚至我为Cursor花的钱比Windsurf多。我认为它们走的是两条不同的路线：Cursor 的眼光更大，瞄准的人群更广，包括很多完全不懂编程的人，上手门槛低；而 Windsurf 更适合那些想要深入理解代码，并构建可维护系统的开发者。 ✅独特的产品哲学 - 最强大稳定的上下文引擎，是人 在长期使用后，我觉得 Windsurf 隐藏着一个独到的产品哲学：最强大的上下文引擎，应该是用户本身。好的 AI 产品， 不仅帮助用户完成当前时间节点的任务，还能在更长的时间维度上，提升用户的某种能力。 Windsurf通过三个feature，让我始终对代码库有一种“熟悉感”，虽然不如程序员那般深刻，但能明显提升AI coding的准确性 (像宝玉老师 @dotey 说的那样，对代码结构与原理理解越深的人，AI赋能的效率越高)： - DeepWiki - CodeMap - Agent对话框中直接引用代码片段 随着对代码库的深入了解，会很自然的移除很多噪音，直达重要的上下文，然后命令 Windsurf Cascade 像一把手术刀一样进行修改代码 (尤其是配合最近出的 Fast Context )。 1️⃣ DeepWiki 这个工具应该很多人都用过，尤其是它的网页版。这个功能简直是心头好，简单来说，把鼠标悬停在变量、函数或类上时，它会根据当前代码库自动解释其在上下文中的含义。 比较厉害的地方是，它甚至能解释引用的某个库中的类或方法。面对不熟悉的代码库或者曾经写的代码，或AI新加的代码，Deep Wiki能快速帮我理清思路，哪些变量、函数或类是干什么的。让我始终能通过自然语言了解项目的大致工作原理，即使有些具体实现我暂时无法理解。 Deep Wiki速度快，而且能让我的注意力始终在代码上，而不是在对话框和代码间反复切换，我也不用花精力去写提示词，比如“请你解释...”。 总之，速度+简单的操作，让我能始终保持在心流中。而且 Deep Wiki 的结果可以直接加入到 Agent 对话框中，提供非常精准的上下文。 2️⃣ CodeMap 这是最近推出的功能，之前在预览版中使用过，很是喜欢。Code Maps 会根据我的问题可视化调用链路，告诉我某个功能的完整实现路径是什么样的。我可以跟着调用链一步步了解代码的运行机制，至少留个印象。多用用就会发现很多链路是相似的，此时让AI总结下方法论，基于已有的印象，曾经无法理解的方法，现在就清楚了。 也可以将这个链路添加到 Agent 上下文中，让 Agent 针对这个链路进行代码审查、新增功能或查找 bug。这样上下文会更加准确。 3️⃣ Agent对话框中直接引用代码片段 在 Cursor 中，当想引用某些代码片段时，只能将它加入到对话框顶部作为上下文，但无法在对话内容中直接引用。而在 Windsurf 中，我可以在对话框中直接引用和描述选中的代码片段，就像一把手术刀一样精准。只要有明确的修改想法，用这个功能就能实现精确控制 当然对于Windsurf也有很多想吐槽的地方，比如上下文窗口不够大。如果是纯Vibe Coding，可能上下文就没那么够用了。我觉得这受限于Windsurf目前的按积分收费的机制，让其得不得压缩上下文来控制成本。不过你如果上下文提供较为相关，Windsurf Cascade的上下文还是完全够用的。 对于非程序员背景的我来说，Vibe Coding 很迷人，但碰壁也很快，想要走得更远，了解原理是无法规避的关卡。感谢Windsurf加快了我的进程。

Codemaps are now available to all users! I've been super excited for this to launch as it's my favorite feature in @windsurf. Solving context is one of the hardest problems in agentic coding and Codemaps gets us one step closer You can now: - generate codemaps from many places in Windsurf: in the codemaps tab, within any file, from a Cascade conversation - ask questions about your codemaps - step through rich diagrams in the codemaps - bring any codemap as additional context to Cascade

Codemaps can also be taken literally - visual system diagrams with two way linking strongly grounded in the codebase. we heard this feedback from Aiden but also our biggest Fortune 500 customers trying to wrangle their codebases x.com/aidenybai/status/19850… see it here: windsurf.com/codemaps/87532a… in the toggle in sidebar

Full guide and blogpost: cognition.ai/blog/codemaps Try it out in latest Windsurf: windsurf.com/download or see/try Codemaps online: windsurf.com/codemaps/87532a… and in DeepWiki

Codemaps is now live! We have found that by far the largest constraint on your ability code, whether manually or with agents, is your ability to understand the code you are working with. When you move fast with low understanding, you get slop. When you move fast and scale understanding with AI, you can truly scale your productive output. Fight slop with Codemaps.

Devin now has full computer use capabilities and can share screen recordings. You can control desktop apps, build and QA mobile apps, and automate tedious work. Here are some examples that blew our team away: 1. Making a desktop game

Reload Windsurf or download the latest version to try it out: windsurf.com/download/editor

Our blogpost: cognition.ai/blog/swe-1-5

Read our full research announcement:

Today we’re releasing SWE-1.5, our fast agent model. It achieves near-SOTA coding performance while setting a new standard for speed. Now available in @windsurf.

Proud to be joining @github's new Agent HQ as launch partners. More choices and more ways to try Devin, coming later this year!

Jupyter Notebooks are now supported in Cascade! 📕

Make sure to update Windsurf to the latest version to try it out! windsurf.com/download/editor

You can now try a new stealth model in Windsurf. Falcon Alpha is a powerful agentic model designed for speed. We're excited to hear what you think ⚡️

Read @cognition's announcement below:

Download Windsurf to use SWE-grep-mini alongside your favorite frontier agents. Rolling out gradually to users in the coming days. windsurf.com/download/

Feel the speed of SWE-grep-mini in our interactive playground: playground.cognition.ai

We trained a first-of-its-kind family of models: SWE-grep and SWE-grep-mini. Designed for fast agentic search (>2,800 TPS), surface the right files to your coding agent 20x faster. Now rolling out gradually to Windsurf users via the Fast Context subagent.