Skywork-R1V3-38B

Skywork-R1V3-38B, released by Kunlun Tech's Skywork AI team in July 2025, is an open-source multimodal reasoning model trained using reinforcement learning to advance visual reasoning capabilities. It contributes to the research community's understanding of RL-based training for vision-language models — demonstrating that the RL approach that has driven advances in text reasoning is also an effective method for improving visual understanding in openly available, practically deployable models.