Your prompts, kept working — as the models change.
Paste a prompt below. We diagnose it, rewrite it for each major model, and run the evidence side by side — then we keep watching, and tell you when a model release changes the answer.
I mainly use
Free · 3 tests a day · runs on fast models, clearly labeled · your prompt is never stored without asking
From this month's notebook
Twenty everyday tasks, re-tested on every release.
battery № 47 · run 2026-08-25 · methodology public¹
- Marketing copyClaude takes the lead — cleaner hooks, less boilerplate on 4 of 5 taskschanged
- Long-doc summaryno change — GPT holds for the third release runningholds
- Data extractionGemini, notably improved — if you gave up on it in spring, retry+18%
What you get
Every test ends in a report you can keep.
Diagnosis, a rewrite tuned per model, live before/after proof, and this month's verdict on which model suits your task.
See a sample report →Keep this working
Know when the answer changes.
Models ship monthly. When a release changes the best choice for your task, you'll know within 48 hours — with the re-tuned prompt attached. Nothing else, no weekly filler.