LLM-as-a-Judge on RewardBench
92.9AccuracyQwen3-Next-80B-A3B-Thinking
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen3-Next-80B-A3B-ThinkingParameters=80B, Active Parameters=3B, Mode=Thinking2026.01 | 92.9 | |
| Qwen3-30B-A3B-Thinking-2507Parameters=30B, Active Parameters=3B, Mode=Thinking, Version=25072026.01 | 92.01 | |
| DeepSeek-R12026.01 | 91.18 | |
| QwQ-32BParameters=32B2026.01 | 91.05 | |
| Qwen3-30B-A3B-Instruct-2507Parameters=30B, Active Parameters=3B, Mode=Instruct, Version=25072026.01 | 89.88 | |
| DeepSeek-V32026.01 | 89.74 | |
| Qwen2.5-32B-InstructParameters=32B, Mode=Instruct2026.01 | 89.31 | |
| Qwen3-Next-80B-A3B-InstructParameters=80B, Active Parameters=3B, Mode=Instruct2026.01 | 88.96 | |
| GeniiBackbone=Qwen2.5-14B-Instruct, Evaluation Setup=Same Model Family2025.10 | 82.48 | |
| VanillaBackbone=Qwen2.5-14B-Instruct, Evaluation Setup=Same Model Family2025.10 | 80.8 | |
| Judgment SFTBackbone=Gemma2-9B-Instruct, Evaluation Setup=Different Model Families2025.10 | 73.77 | |
| GeniiBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 73.6 | |
| GeniiBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Same Model Family2025.10 | 72.63 | |
| Judgment SFTBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 71.48 | |
| Judgment SFTBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 71.06 | |
| Self-ConsistencyBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 69.57 | |
| VanillaBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 69.41 | |
| VanillaBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Same Model Family2025.10 | 69.41 | |
| GeniiBackbone=Gemma2-9B-Instruct, Evaluation Setup=Different Model Families2025.10 | 67.17 | |
| RepromptBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 66.6 | |
| Self-ConsistencyBackbone=Gemma2-9B-Instruct, Evaluation Setup=Different Model Families2025.10 | 66.36 | |
| Long ReasoningBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 66.3 | |
| VanillaBackbone=Gemma2-9B-Instruct, Evaluation Setup=Different Model Families2025.10 | 64.66 | |
| Long ReasoningBackbone=Qwen2.5-7B-Instruct, Evaluation Setup=Different Model Families2025.10 | 63.95 | |
| RepromptBackbone=Gemma2-9B-Instruct, Evaluation Setup=Different Model Families2025.10 | 60.4 | |
| VanillaBackbone=Qwen2.5-3B-Instruct, Evaluation Setup=Same Model Family2025.10 | 57.89 | |
| GeniiBackbone=Qwen2.5-3B-Instruct, Evaluation Setup=Same Model Family2025.10 | 56.65 | |
| GeniiBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 51.15 | |
| Self-ConsistencyBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 50.5 | |
| VanillaBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 50.28 | |
| RepromptBackbone=Llama3.1-8B-Instruct, Evaluation Setup=Different Model Families2025.10 | 40.2 |