Is Ajax everything it claims to be?

How it was tested

Every model got the same questions in the same order, at temperature 0 with the same seed. Each one was unloaded from the graphics card before the next ran, so neither inherited the other's warm start. Speed is the median of three timed runs per question. Tool questions are checked by rule: the right tool, with the right details. Everything else was marked by Claude Sonnet 5.5, a model from outside the Qwen family, with its reason under each answer. A judge can be wrong, which is why every answer and every reason is printed here for you to check.

What this does not cover: one PC is one PC, and ten questions per topic is a sample, not a census. Requests for dangerous instructions were not tested; that is not what RigMatch measures.

RigMatch is a free, independent app. This report is not affiliated with or endorsed by the Odysseus project, Ajax's makers, or PewDiePie.