CLEM Clean Up: leaderboard
CLEM text release v3.0: models coordinate a cleanup task under conversational interaction rules. Scores combine protocol-valid play and conditional task quality, higher is better. Headline preserves the native published product of macro-average played percentage and macro-average conditional quality; per-game scores multiply source-published played and quality components without extra rounding. Source temperature, player configuration and exact evidenced model/checkpoint settings are retained. Undefined conditional quality remains missing rather than imputed zero.
Metric: Game Clemscore (%), text v3.0; published played × quality / 100. Source: huggingface.co. Saturation forecast: Estimated already saturated. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (High) | 100 | #110 (GPT-5.2) |
| 2 | GPT-5.2 (Medium) | 99.94 | #110 (GPT-5.2) |
| 3 | Kimi K2 (Thinking) | 98.04 | #243 (Kimi K2) |
| 4 | Gemini 3 Flash (Preview) | 97.25 | #82 |
| 5 | GLM-4.7 | 87.06 | #192 |
| 6 | GLM-4.6 | 83.29 | #252 |
| 7 | MiniMax-M2.5 | 74.22 | #303 |
| 8 | Qwen 3 Max | 72.04 | #208 |
| 9 | DeepSeek V3.2 (Non-reasoning) | 70.55 | #205 (DeepSeek V3.2) |
| 10 | GLM-5 (Non-reasoning) | 70.41 | #141 (GLM-5) |
| 11 | DeepSeek V3.2 | 68.43 | #205 |
| 12 | Kimi K2.5 (Non-reasoning) | 61.59 | #144 (Kimi K2.5) |
| 13 | Qwen 2.5 72B Instruct | 41.09 | #454 |
| 14 | Qwen 3 Next 80B A3B Instruct | 40.38 | #368 |
| 15 | Llama 3.1 70B Instruct | 38.05 | #561 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clem-clean-up · How It Works · Data refreshed daily, snapshot 2026-10-20.