WuYuEval - Foundation Module - Macro-F1: leaderboard
Macro-F1 on the Foundation Module of WuYuEval (arXiv 2608.07529v1, Appendix Table 7): 4,590 audited closed-ended multiple-choice solid-waste-management questions across six task types and eight domain categories. Percent; higher is better; standard and thinking modes of the same model are separate rows.
Metric: Macro-F1 (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 95.73 | #87 |
| 2 | Gemini 3 Pro (Preview) | 95.53 | #71 |
| 3 | GLM-4.6 (Non-reasoning) | 93.54 | #265 (GLM-4.6) |
| 4 | DeepSeek V3.2 (Thinking) | 93.51 | #210 (DeepSeek V3.2) |
| 5 | Qwen 3 Max | 93.09 | #213 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 92.66 | #210 (DeepSeek V3.2) |
| 7 | Qwen 3 235B A22B (Thinking) | 92.54 | #323 (Qwen 3 235B A22B) |
| 8 | Kimi K2 (Thinking) | 92.23 | #252 (Kimi K2) |
| 9 | GPT-5.2 | 91.36 | #115 |
| 10 | Qwen 3 235B A22B Instruct | 90.93 | #336 |
| 11 | Qwen 3 32B (Thinking) | 89.58 | #449 (Qwen 3 32B) |
| 12 | Qwen 3 32B (Non-reasoning) | 89.18 | #449 (Qwen 3 32B) |
| 13 | Qwen 3 14B (Reasoning) | 87.7 | #542 (Qwen 3 14B) |
| 14 | GLM-4.6 (Thinking) | 86.53 | #265 (GLM-4.6) |
| 15 | Gemma 3 27B (IT) | 84.73 | #528 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=wuyueval-foundation-module-macro-f1 · How It Works · Data refreshed daily, snapshot 2026-11-05.