Skywork releases Reward-V2, eight reward models from 0.6B to 8B
For context: Llama and Qwen3-based preference scorers trained on about 26 million pairs. The 8B card reports 96.4 on RewardBench v1.
The AI that actually shipped
For context: Llama and Qwen3-based preference scorers trained on about 26 million pairs. The 8B card reports 96.4 on RewardBench v1.