Geospatial Text Generation and Spatial Reasoning on GeoHeight-Bench
12.22Error Rate (ER)InternVL2-4B
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| InternVL2-4BCategory=Open-source LMMs2026.03 | 12.22 | 10.27 | 22 | 7.75 | 21.9 | 46.44 | 34.04 | 22.09 | |
| InternVL2-8BCategory=Open-source LMMs2026.03 | 14.01 | 3.9 | 13.51 | 5.41 | 17.06 | 32.62 | 38.63 | 17.88 | |
| Ministral3-8BCategory=Open-source LMMs2026.03 | 17.67 | 6.17 | 5.28 | 4.05 | 15.77 | 33.14 | 8.15 | 12.89 | |
| InternVL2-2BCategory=Open-source LMMs2026.03 | 17.78 | 5.77 | 18.8 | 5.27 | 15.75 | 39.19 | 10.94 | 16.22 | |
| GPT-4oCategory=Closed-source LMMs2026.03 | 19.4 | 7.02 | 12.2 | 8.18 | 24.89 | 45.34 | 34.23 | 21.61 | |
| LLaVA-NeXT-7B-MistralCategory=Open-source LMMs2026.03 | 22.78 | 1.97 | 7.18 | 5.78 | 19.33 | 44.99 | 10.76 | 16.11 | |
| Gemini-2.5-FlashCategory=Closed-source LMMs2026.03 | 28.11 | 11.45 | 29.82 | 22.57 | 27.11 | 37.3 | 42.74 | 28.45 | |
| Pixtral-12BCategory=Open-source LMMs2026.03 | 30.6 | 7.77 | 9.71 | 5.52 | 20.17 | 40.3 | 12.23 | 18.04 | |
| LLaVA-NeXT-7B-VicunaCategory=Open-source LMMs2026.03 | 34.07 | 9.16 | 9.63 | 5.83 | 18.06 | 44.54 | 10.12 | 18.78 | |
| LLaVA1.5-7BCategory=Open-source LMMs2026.03 | 36.72 | 9 | 7.83 | 5.41 | 27.59 | 44.73 | 10.03 | 20.19 | |
| LLaVA-NeXT-34BCategory=Open-source LMMs2026.03 | 51.37 | 7.85 | 10.43 | 6.53 | 20.17 | 45.84 | 11 | 21.88 | |
| LLaVA1.5-13BCategory=Open-source LMMs2026.03 | 53.46 | 9.48 | 7.07 | 5.14 | 22.96 | 40.76 | 10.25 | 21.3 | |
| LLaVA-NeXT-13BCategory=Open-source LMMs2026.03 | 55.64 | 9.48 | 9.27 | 7.33 | 20.4 | 44.98 | 10.42 | 22.5 | |
| LISA-7B (FT)Category=Height-aware LMMs, Fine-tuned=true2026.03 | 56.58 | 17.53 | 25.75 | 26.25 | 16.12 | 47.54 | 32.31 | 31.73 | |
| GeoHeightChatCategory=Height-aware LMMs2026.03 | 59.98 | 23.36 | 42.19 | 49.09 | 26.16 | 55.24 | 52.97 | 44.14 |