Do LLMs Build World Models From Text? A Multil: leaderboard
Whether large language models (LLMs) construct internal spatial world models from pure-text descriptions remains contested, and whether such capabilities transfer across languages has not been systematically studied. We introduce MentalMap, a multilingual diagnostic benchmark with a six-level capability hierarchy (L0-L5) spanning atomic spatial facts to generative world-graph construction, togethe
Metric: L3 (viewpoint) (self-reported). Source: benchmarklist.com. Saturation forecast: Around March 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 32.4 | #343 |
| 2 | Gemma 3 27B (IT) | 28.7 | #523 |
| 3 | DeepSeek V4 Flash | 26.4 | #137 |
| 4 | Gemini 2.5 Flash | 21.3 | #245 |
| 5 | Qwen 2.5 7B Instruct | 18.9 | #862 |
| 6 | Falcon3-10B-Instruct | 16.4 | #1157 |
| 7 | Qwen 2.5 32B Instruct | 14.7 | #498 |
| 8 | Qwen 3.5 Flash | 14.2 | #232 |
| 9 | Gemma 2 9B (IT) | 13.7 | #788 |
| 10 | Llama 3.1 8B Instruct | 13.3 | #1035 |
| 11 | Mistral 7B Instruct (v0.3) | 1 | #1274 |
| 12 | SmolLM2-1.7B-Instruct | 0 | #1562 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=do-llms-build-world-models-from-text-a-multil · How It Works · Data refreshed daily, snapshot 2026-10-20.