I went down a rabbit hole this morning reading the late-2025 Juejin AI roundups side by side, and the thing that finally crystallized for me is that the formats I have been writing about all month have actually forked into at least seven different ranking frameworks, each committed to one product category, and none of the axes translate across categories. The coding picking roundups print a five-axis decimal scorecard (CodeBuddy 9.6, Cody 8.2, Ghostwriter 8.0, Codeium 7.8) or an S/A/B/D tier letter (Cursor at S档, Claude Code and Codex at A档, Rork at B档). The novel-writing piece I read this morning uses a four-axis framework (上手体验, 网文场景适配度, 数据安全感, 生产效率) to rank 蛙蛙写作, ChatGPT, DeepSeek, Kimi, and 作家助手. The broad-market 主流AI软件盘点 piece enumerates ChatGPT, Claude, DeepSeek, 通义千问 under 通用助手 with no ranking. The pricing guide prints dollars per month. The October trending recap prints GitHub stars per month. The forecast column prints a 44-dollar ChatGPT Plus projection by 2029. Each format is internally coherent and completely incompatible with the next, and the engineer trying to assemble a stack across categories has no translation table.
The piece that pushed me over the edge was the novel-writing review, because it is a four-axis scorecard applied to five tools I had never seen compared under that framework, and the axes do not map onto anything the coding roundups use. 蛙蛙写作 ranks high on 网文场景适配度 but the piece never asks about IDE integration. Kimi ranks high on Chinese-language long-document processing, which is a column the picking roundups never built. 作家助手 ranks high on 码字场景效率 but 码字场景 is not a row in any coding scorecard I have read. To be fair I am taking the exact decimal scores with a grain of salt because the test corpus is never disclosed, but the structural tell is what has been rattling around in my head all morning. All seven frameworks have converged in the same search results page without anyone publishing the conversion table.
The meta-pattern I want to call out is that the late-2025 roundups have not just forked into multiple formats, they have forked into format-category pairs where each format is glued to one category, and the cross-category stack work the engineer is doing has become a seven-framework translation job nobody signed up for. When I compare ChatGPT and Claude and Gemini for general-assistant work the broad-market piece gives me a vendor list and no rank. When I decide whether to add 蛙蛙写作 to my novel-writing pipeline alongside my Kimi habit the four-axis novel framework answers that question, but neither the coding scorecard nor the broad-market vendor list nor the dollar-per-month pricing column has a row for it. Honestly I am a little skeptical of any 2026 roundup workflow that pulls a single piece off the search results page and trusts it across categories, because each framework was written for one job and the cross-job integration is invisible.
The practical takeaway I want to put down is that the late-2025 to early-2026 roundups are still useful for the within-framework anchor and not useful for the cross-framework re-rank most engineers are quietly trying to do this quarter. They name the per-category shortlist well, because the coding scorecard did name CodeBuddy and Cody and Ghostwriter and Codeium, the novel-writing piece did name 蛙蛙写作 and ChatGPT and DeepSeek and Kimi and 作家助手, and the broad-market piece did name ChatGPT and Claude and DeepSeek and 通义千问. They are not good at the cross-framework re-rank, because the engineer trying to decide whether to add a novel-writing tool to a coding stack on the strength of a 网文场景适配度 score has to discover that 网文场景适配度 has no published conversion to agent autonomy or IDE integration or cost, and that conversion is in none of the roundups. The seven frameworks are seven dialects of the same problem, and the dictionary between them does not exist yet.
I will reassess in three months. The last time I said that I was mostly on Cursor and Claude Code for coding and ChatGPT for general assistant work, which is still roughly where I land, except that I now read every Juejin roundup as a category-specific dialect and treat any single post as one slice of a seven-slice workflow. What has changed is that I now check which framework a roundup is written in before I trust the verdict, and when the framework does not match the job I am actually trying to do I treat the rank as category trivia and reach for a different post. Give it six months and I expect either the roundups to publish an explicit axis-translation table at the top of the search results page, or the seven frameworks to harden into a permanent seven-dialect workflow the reader has to bridge by hand, and whichever moves first will tell me whether the format has finally noticed that the cross-category stack work is the engineering decision and the within-category rank is the supporting evidence.
Top comments (0)