I went down a rabbit hole this morning reading the late-2025 Juejin picking roundups side by side, and the mobile-app tier snuck into a column I do not think it belongs in. The 编程工具评测 piece ranked Cursor at S档, Claude Code and Codex at A档, then slid Rork, VibeCode App, and Anything — three mobile-app generators — into the same list, and put Lovable and v0 in a frontend-component row that shares the scorecard with GitHub Copilot and Codeium. The AI开发工具最新排行 piece crowned Cursor at 综合能力第一 with a 20元/月 price, Codeium as 免费, and V0.dev as the Vercel-integrated UI specialist, then folded Lovable, BlackBox AI, Replit, and CodeWhisperer into a single ranked list with no metric stated. After putting the mobile tier next to the editor-loop tier next to the component-tier next to the terminal-agent tier, my honest take is that the picking roundups now mix four artifact categories into one column, and the verdict is collapsing across categories that should not share a ranking.
The piece that crystallized it for me opens with Cursor at S档闭眼选, gives Codex an A档 for fast GPT updates, gives Claude Code an A档 while noting that 代理能力 is no longer ahead of Cursor, then slides Lovable and v0 into the same tier because 一句话生成 covers the same audience as Cursor's Cmd+K, then drops Rork, VibeCode App, and Anything into B档 because they generate iOS and Android via Expo. To be fair I would take the exact tier boundaries with a grain of salt because vendors have shuffled bundles since, but the structural tell is what has been rattling around in my head all morning. Cursor, Codex, Claude Code, v0, Lovable, Rork, VibeCode App, and Anything are all in the same ordered list, and the list never names its sort column.
The meta-pattern I want to put down is that the late-2025 picking roundups are now mixing editor-IDE products, terminal-agent products, mobile-app generators, and component-tier frontend tools into one verdict column without acknowledging that the artifact each one ships is the deciding factor. Cursor and Claude Code leave a reviewable git diff with tests and terminal evidence. Codex leaves a patch or a ChatGPT plan upgrade. v0 leaves a React component pointed at Vercel. Lovable leaves a hosted web prototype. Rork and VibeCode App leave an Expo-built iOS and Android pair. Replit leaves a running hosted prototype. Those handoff objects have completely different lifetimes, verification costs, and review loops, and ranking them in one column is like ranking PostgreSQL, GitHub Actions, and a senior backend engineer because all three can help ship a backend. The S that puts Cursor on top and the B that puts Rork below it look like a verdict on quality, but they are really a verdict on four different jobs the roundup never separated out.
Honestly I am a little skeptical of any 2026 picking roundup that prints a single S/A/B tier letter or a single decimal score across these four artifact categories, because the letter is really telling me four different things depending on which author wrote it. The 综合能力第一 ranking of Cursor is sorting on editor-loop fluency, the 性价比之王 ranking of Codeium is sorting on 免费 versus 20元/月 pricing, the V0.dev as UI specialist ranking is sorting on Vercel deployment, and the Lovable ranking is sorting on prototype speed. None of those axes is the same. My gut says the format has not yet noticed that the row it is scoring no longer maps onto a single shape.
The practical takeaway is that the picking roundups are still useful for the within-category anchor and not useful for the cross-category re-rank most engineers are quietly trying to do this quarter. They are good at telling me Cursor, Claude Code, Codex, GitHub Copilot, and Codeium are the five editor-loop tools worth evaluating, and that v0 and Lovable are the two component-tier tools worth evaluating, and that Rork, VibeCode App, and Anything are the three mobile-app generators worth evaluating. They are not good at telling me which one to pick when the job requires a reviewable diff versus a deployed mobile prototype versus a hosted web app, because those three jobs do not share a verdict column. The S letter and the five-axis decimal both compress four artifact categories into one cell, and the cell no longer carries the information the reader needs.
I will reassess in three months. For now I am still mostly on Cursor and Claude Code for repository work, ChatGPT for open-ended exploration, GitHub Copilot when GitHub and VS Code integration removes friction, and v0 when the deliverable is a component rather than a feature, and I am explicitly leaving Lovable, Rork, VibeCode App, and Anything out of my stack until the job genuinely asks for a hosted prototype or a mobile pair. Give it six months and I expect either the picking roundups to bolt on an explicit artifact category column — editor IDE, terminal agent, hosted prototype, component tier, mobile pair — or the tier letter and the decimal scorecard to harden into a permanent cross-category verdict the reader has to decompose by hand, and whichever moves first will tell me whether the format has finally noticed that the row it is ranking is no longer one shape.
Top comments (0)