In July 2025, Skywork announced the release of its second-generation reward model series, Skywork-Reward-V2, expanding on its successful open-sourced models from September 2024. These models, available on HuggingFace and GitHub, have been downloaded over 750,000 times, achieving top rankings in seven evaluation benchmarks. The Skywork-Reward-V2 features eight models with parameters ranging from 600 million to 8 billion, demonstrating exceptional performance in tasks requiring human-aligned preferences. The underlying innovation, Skywork-SynPref-40M, is a hybrid dataset containing 40 million preference pairs, developed through a unique human-machine collaboration for data screening. This approach enhances the model’s efficiency and effectiveness, showcasing significant advancements in Reinforcement Learning from Human Feedback (RLHF). With strong generalization capabilities and superior performance compared to larger models, Skywork forecasts a pivotal role for unified reward systems in AI’s future infrastructure, guiding intelligent systems to align with human values. Visit Skywork.AI to explore the latest developments.