LLM-as-a-Judge Robustness to Adversarial Attacks on RobustJudge
-0.109None Condition ScoreQwen3-Next-80B-A3B-Instruct
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-Next-80B-A3B-InstructSize=80B, Variant=Instruct, MoE=A3B2026.01 | -0.109 | — | -0.045 | -0.044 | 0.198 | -0.023 | -0.051 | 0.353 | 0.759 | -0.806 | 0.026 | |
| Qwen3-30B-A3B-Instruct-2507Size=30B, Variant=Instruct, MoE=A3B, Version=25072026.01 | -0.129 | — | -0.076 | -0.045 | 0.047 | 0.042 | -0.024 | 0.273 | 0.859 | -0.532 | 0.046 | |
| Qwen2.5-32B-InstructSize=32B, Variant=Instruct2026.01 | -0.213 | — | -0.65 | -0.156 | 0.517 | -0.172 | -0.18 | -0.146 | 0.406 | -0.65 | -0.138 | |
| DeepSeek-V32026.01 | -0.259 | — | -0.217 | -0.19 | 0.51 | -0.139 | -0.197 | -0.043 | 0.35 | -0.695 | -0.098 | |
| QwQ-32BSize=32B2026.01 | -0.316 | — | -0.652 | -0.261 | 0.517 | -0.26 | -0.268 | 0.508 | 0.535 | -0.652 | -0.094 | |
| Qwen3-Next-80B-A3B-ThinkingSize=80B, Variant=Thinking, MoE=A3B2026.01 | -0.383 | — | -0.401 | -0.312 | 0.461 | -0.277 | -0.439 | 0.466 | -0.009 | -0.815 | -0.19 | |
| Qwen3-30B-A3B-Thinking-2507Size=30B, Variant=Thinking, MoE=A3B, Version=25072026.01 | -0.412 | — | -0.336 | -0.321 | -0.316 | -0.297 | -0.433 | 0.17 | 0.511 | -0.702 | -0.237 | |
| DeepSeek-R12026.01 | -0.434 | — | -0.379 | -0.357 | 0.366 | -0.326 | -0.375 | -0.265 | 0.882 | -0.734 | -0.18 |