Tool
Model picker quiz.
Five questions, one recommendation — with the reasoning printed, because a picker you can't audit is just an ad.
How this works
Hard filters first (open weights, minimum context). Then transparent points: League Score, a bonus for a matching strength, division rank for coding/reasoning, a value term scaled by your budget answer, and a speed term from published tokens/sec where available. The same rules are printed here — no sponsored placements, ever.