Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning
Open paper
Benchmark
Model
Rank
Results
mathematical-reasoning-on-aime-2025
G2RL (Qwen3-4B-Base)
–
Accuracy: 20.1
mathematical-reasoning-on-aime24
G2RL (Qwen3-4B-Base)
–
Acc: 19.9
mathematical-reasoning-on-math-500
G2RL (Qwen3-4B-Base)
–
Accuracy: 80.8
Rank counts only results with a code link.