Grok-2 mini

accuracy • Self-reported

pass@1 Method determination probability correct answer model with first attempts at n attempts. This metric accounts for probability, with which model correct answer at use, at that at us is n answers model on one and that indeed question. for pass@1 on "majority ": correct answer should more "" (probability), than incorrect answer. this can be with help sample with In case two possible answers, pass@1 shows probability that, that model correct answer in on basis correct answers in n • Self-reported

maj@1 AI: on which indicates majority from 1 output. For models, to with help options, sometimes useful use In case maj@1 we simply one output and then, that he This useful how basic method for comparison with other methods • Self-reported

accuracy • Self-reported

accuracy • Self-reported

accuracy • Self-reported

accuracy • Self-reported

accuracy • Self-reported