Kunlun Wanwei has recently unveiled its groundbreaking open-source model, Skywork-R1V3.0, which is setting new benchmarks in multimodal reasoning capabilities. This state-of-the-art model rivals the proficiency of novice human experts, boasting significant advancements in complex logic modeling and interdisciplinary knowledge generalization. Skywork-R1V3.0 employs reinforcement learning strategies during training, leading to remarkable progress in these areas.

Building on the foundation of its predecessor, Skywork-R1V2.0, Skywork-R1V3.0 leverages high-quality distillation data and rejection sampling techniques to construct a robust multimodal reasoning training set. The model transcends mere text processing, incorporating image handling capabilities and significantly enhancing its reasoning abilities between images and text.

Skywork-R1V3.0's training is impressively efficient, relying on approximately 12,000 supervised fine-tuning samples and 13,000 reinforcement learning samples. This demonstrates the unique advantage of "small data, big power." In the authoritative multimodal evaluation, MMMU, Skywork-R1V3.0 leads with a score of 76.0, outperforming closed-source models like Claude-3.7-Sonnet (75.0) and GPT-4.5 (74.4), showcasing its exceptional cross-modal understanding capabilities.

In specific application scenarios, Skywork-R1V3.0 excels across various domains, including physics, logic, and mathematical reasoning. In physics reasoning evaluations, the model achieves the best open-source scores of 52.8 and 31.5, demonstrating its understanding of complex physical problems. In logical reasoning tests, Skywork-R1V3.0 also scores an impressive 59.7.

The model's prowess in mathematical reasoning is equally noteworthy, with outstanding scores of 77.1, 59.6, and 52.6 in MathVista, MathVerse, and MathVision evaluations, respectively, significantly outperforming other open-source models. These stellar performances position Skywork-R1V3.0 as a formidable contender in the current open-source multimodal reasoning landscape.

The release of Skywork-R1V3.0 signifies a new pinnacle in multimodal reasoning technology. Its powerful performance and open-source nature will greatly facilitate further advancements in AI technology, ushering in a new era of innovation and progress.