I went down a rabbit hole this morning reading the late-2025 Juejin AI tool coverage side by side — the 2025 coding tools review with the S/A/B/D tier ranking, the November frontend picking guide with the 按预算选择 structure, the December authoritative eight-mainstream coding ranking with the 9.6/8.2/7.8 decimal scorecards, and the 2025 Chinese AI tools directory with seven categorical sections plus a 神马中转 API pivot — and the thing that finally crystallized for me is that the picking-format itself has forked into at least three incompatible sub-formats on the same page, and the engineer who reads two picks in the same week can get two different rankings on the same tool without either one being obviously wrong.
The piece that pushed me over the edge was reading the S/A/B/D tier ranking piece next to the decimal scorecard piece and noticing that they share almost no product names and use completely different ranking currencies. The S/A/B/D piece put Cursor at S-tier, v0 at S/A boundary, Claude Code at A-tier, Replit and Chef at B-tier, and explicitly skipped C-tier with a footnote saying the C-tier did not have any meaningful candidates this year. The decimal scorecard piece put CodeBuddy at 9.6, Sourcegraph Cody at 8.2, Replit Ghostwriter at 8.0, Codeium at 7.8, Tabnine at 7.6, CodeWhisperer at 7.5, JetBrains AI Assistant at 7.4, and Blackbox at 7.2 on a five-axis grid. To be fair I am taking the exact decimal scores with a grain of salt because both pieces were published in the last quarter of 2025 and the test corpora are the authors' own workloads, but the structural disjointness is what has been rattling around in my head all morning. Two pieces from the same genre, same month, same language, with the same headline framing, and the verdict-overlap is almost zero.
The meta-pattern I want to call out is that the late-2025 Juejin picking-format has split into three sub-formats that no longer agree on what a pick looks like. The tier-grading sub-format puts tools into S/A/B/D buckets with no C and with v0 sitting ambiguously on the S/A boundary. The decimal-scorecard sub-format prints a 9.6/8.2/7.8 grid across five axes with a separate row for each tool. The budget-tier sub-format in the frontend guide ranks tools by how much monthly spend the engineer can absorb (预算充足, 预算有限, 无预算), with GitHub Copilot at ten dollars per month anchoring the cheap tier and Cursor at twenty dollars per month anchoring the comfortable tier. Honestly I am a little skeptical of any 2026 picking roundup that treats a single piece as a single source of truth, because the three sub-formats are really telling me the picking genre has forked internally and the engineer who reads one picking post is reading a third of the picking story.
The practical takeaway I want to put down is that the late-2025 to early-2026 Juejin picking roundups are still useful for three narrow jobs and not useful for the fourth job most engineers are quietly trying to do. The tier-grading sub-format is good at the quick-bucketing job, because the S/A/B/D piece gives a busy engineer a one-glance verdict that survives a skim read. The decimal-scorecard sub-format is good at the within-axis spread job, because the spread between CodeBuddy at 9.6 and Blackbox at 7.2 is exactly the kind of decision a team lead needs to make between two autocomplete candidates. The budget-tier sub-format is good at the monthly-bill-matching job, because the 按预算选择 structure tells an engineer with a tight budget to start with Codeium and V0 instead of Cursor and GitHub Copilot. None of them are good at the cross-sub-format consensus job, because the engineer has to take the tier verdict from one piece and the decimal verdict from another and the budget verdict from a third, and the three verdicts on the same tool can land in three different buckets without any of them being dishonest.
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 everything else, which is still roughly where I land, except the picking-format split is now making the within-IDE-tooling decision harder rather than easier. What has changed is that I now read the tier-grading pieces for the quick-bucketing signal, the decimal-scorecard pieces for the within-axis spread, and the budget-tier pieces for the monthly-bill-matching verdict, and I treat any single picking roundup as one slice of a three-slice workflow rather than as a verdict. Give it six months and I expect either the tier-grading sub-format to publish a decimal scorecard alongside each tier or the decimal-scorecard sub-format to publish an explicit tier alongside each row, and whichever moves first will tell me whether the picking-format is going to collapse back into one currency or harden into a permanent reader-bridging chore.
Top comments (0)