Reward Model Evaluation on Text-Coding (test fraction 0.2)
80.9Pair AccuracyArena BT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Arena BTReward Model=proprietary RM2026.06 | 80.9 | 0.773 | 1 | 0.851 | |
| Arena Avg RewardReward Model=proprietary RM2026.06 | 80.5 | 0.779 | 0.994 | 0.851 | |
| DUALEVALReward Model=proprietary RM2026.06 | 77.8 | 0.959 | 0.878 | 0.845 | |
| Static 2PL2026.06 | 73.1 | 1 | 0.773 | 0.763 | |
| Arena BTReward Model=Skywork-Reward-V2-Qwen3-8B2026.06 | 68 | 0.255 | 0.416 | 0.511 | |
| Arena Avg RewardReward Model=Skywork-Reward-V2-Qwen3-8B2026.06 | 67.3 | 0.197 | 0.311 | 0.432 | |
| DUALEVALReward Model=Skywork-Reward-V2-Qwen3-8B2026.06 | 59.1 | 0.976 | 0.805 | 0.789 |