CLEM TextMapWorld GraphReasoning: leaderboard
CLEM text release v3.0: models reason over the graph structure of a textual map. 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: Around January 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2 (Thinking) | 84.28 | #243 (Kimi K2) |
| 2 | Gemini 3 Flash (Preview) | 83.08 | #82 |
| 3 | Qwen 3 Max | 82.32 | #208 |
| 4 | Kimi K2.5 (Non-reasoning) | 73.16 | #144 (Kimi K2.5) |
| 5 | GLM-5 (Non-reasoning) | 71.99 | #141 (GLM-5) |
| 6 | GLM-4.7 | 69.95 | #192 |
| 7 | DeepSeek V3.2 | 68.47 | #205 |
| 8 | MiniMax-M2.5 | 67.25 | #303 |
| 9 | Qwen 2.5 Coder 32B Instruct | 63.24 | #528 |
| 10 | DeepSeek V3.2 (Non-reasoning) | 61.18 | #205 (DeepSeek V3.2) |
| 11 | Qwen 2.5 72B Instruct | 61.04 | #454 |
| 12 | Qwen 3 Next 80B A3B Instruct | 60.58 | #368 |
| 13 | Llama 3.1 70B Instruct | 58.05 | #561 |
| 14 | GPT-OSS-120B | 58.02 | #341 |
| 15 | GLM-4.6 | 57.42 | #252 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clem-textmapworld-graphreasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.