JustRL: Scaling a 1.5B LLM with a Simple RL Recipe

Benchmark Model Rank Results
mathematical-reasoning-on-olympiadbenchJustRL-Nemotron 1.5B#1Accuracy: 76.59
mathematical-reasoning-on-minerva-mathJustRL-DeepSeek 1.5B#7Accuracy: 51.47
mathematical-reasoning-on-hmmt-feb-2025JustRL-Nemotron 1.5B#19Accuracy: 40.63
mathematical-reasoning-on-math-500JustRL-Nemotron 1.5B#37Accuracy: 94.15
mathematical-reasoning-on-aime24JustRL-Nemotron 1.5B#48Acc: 69.69
mathematical-reasoning-on-aime-2025JustRL-Nemotron 1.5B#65Accuracy: 62.92