WuYuEval - Expert Module - Judge Score: leaderboard
Mean anchor-calibrated LLM-as-a-judge score on WuYuEval's Expert Module of 247 scenario-based open-ended solid-waste-management questions requiring multi-objective optimization, constraint trade-offs and system design (WuYuEval, arXiv 2608.07529v1, Appendix Table 12). 0-1 scale; higher is better.
Metric: Mean anchor-calibrated LLM-as-a-judge score (0-1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B (Thinking) | 0.87 | #323 (Qwen 3 235B A22B) |
| 2 | Qwen 3 Max | 0.87 | #213 |
| 3 | GPT-5.2 | 0.86 | #115 |
| 4 | Kimi K2 (Thinking) | 0.86 | #252 (Kimi K2) |
| 5 | Qwen 3 235B A22B Instruct | 0.85 | #336 |
| 6 | Claude Opus 4.5 | 0.84 | #87 |
| 7 | Gemini 3 Pro (Preview) | 0.82 | #71 |
| 8 | GLM-4.6 (Non-reasoning) | 0.81 | #265 (GLM-4.6) |
| 9 | Qwen 3 32B (Thinking) | 0.8 | #449 (Qwen 3 32B) |
| 10 | Qwen 3 14B (Reasoning) | 0.8 | #542 (Qwen 3 14B) |
| 11 | DeepSeek V3.2 (Thinking) | 0.8 | #210 (DeepSeek V3.2) |
| 12 | Qwen 3 32B (Non-reasoning) | 0.79 | #449 (Qwen 3 32B) |
| 13 | DeepSeek V3.2 (Non-reasoning) | 0.78 | #210 (DeepSeek V3.2) |
| 14 | Qwen 3 14B (Non-reasoning) | 0.78 | #542 (Qwen 3 14B) |
| 15 | Qwen 3 8B (Thinking) | 0.76 | #678 (Qwen 3 8B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=wuyueval-expert-module-judge-score · How It Works · Data refreshed daily, snapshot 2026-11-05.