Step-R1-V-Mini

Multimodal reasoning model with high-precision image perception. Trained via PPO reinforcement learning with verifiable rewards in the image space. Ranked first domestically on MathVision visual reasoning leaderboard.