J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data

Benchmark Model Rank Results
mathematical-reasoning-on-minerva-mathJ-Zero-8B#3Accuracy: 61.76
instruction-following-on-ifevalJ-Zero-4B#15Prompt-level strict-accuracy: 50.65
multiple-choice-question-answering-mcqa-on-supergpqaJ-Zero-8B#20Accuracy: 33.22
mathematical-reasoning-on-olympiadbenchJ-Zero-8B#24Accuracy: 53.12
mathematical-reasoning-on-math-500J-Zero-8B#79Accuracy: 83.4
multi-task-language-understanding-on-mmlu-proJ-Zero-8B#85Accuracy: 63.8
mathematical-reasoning-on-aime24J-Zero-8B#121Acc: 19.58
mathematical-reasoning-on-aime-2025J-Zero-8B#134Accuracy: 15.94