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.
multimodalreasoning

Notes

Released approximately Apr 2025.