Dynamic Cross-View Spatial Intelligence on LinkS2Bench (tiny)
98.3Target Verification (TV)Human Level
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Human LevelCategory=LinkS2Bench (tiny) Perf.2026.04 | 98.3 | 91.3 | 96.7 | 85.8 | 93.6 | 90 | 91.7 | 88.3 | 90 | 93.3 | 95 | 86.7 | 91.7 | 85 | 93.3 | 91.2 | 89.8 | |
| Gemini-3.1-Pro w/ CVAAAlignment Adapter=CVAA, Category=Alignment Adapter Experiments2026.04 | 94.9 | 55.4 | 58.8 | 67.7 | 73.8 | 50 | 47.4 | 53.9 | 50.4 | 36.4 | 36.6 | 43.2 | 38.7 | 48.3 | 60.6 | 66.7 | 58.5 | |
| Qwen3.5-4B_FinetunedCategory=Fine-tuning Experiments2026.04 | 87.2 | 52.6 | 75.9 | 76.4 | 79.8 | 50.8 | 46.3 | 65.4 | 54.2 | 12.5 | 24 | 29.8 | 22.1 | 62.5 | 43.9 | 56.5 | 54.3 | |
| Qwen3.5-9B-Finetuned w/ CVAAAlignment Adapter=CVAA, Category=Alignment Adapter Experiments2026.04 | 87 | 55 | 80 | 85.1 | 84 | 53.2 | 46.7 | 71.7 | 57.2 | 10.4 | 25.5 | 31.9 | 22.6 | 61.2 | 41.3 | 66 | 56.2 | |
| Gemini-3.1-FlashRank=4, Category=Proprietary Models (API)2026.04 | 84.7 | 42.8 | 47 | 68.8 | 66.8 | 35.3 | 40.7 | 49.3 | 41.8 | 15.7 | 25 | 25.7 | 22.1 | 11.3 | 54.7 | 55.7 | 40.6 | |
| Gemini-3.1-ProRank=1, Category=Proprietary Models (API)2026.04 | 84.4 | 51.1 | 57.3 | 63.7 | 68.5 | 44.1 | 44.7 | 51.9 | 46.9 | 29.9 | 30.6 | 47.3 | 35.9 | 44 | 57.3 | 58.1 | 53.1 | |
| Llama-4-ScoutRank=8, Category=Open-source Models2026.04 | 79.6 | 26.6 | 10.4 | 22.5 | 37.5 | 4.1 | 22.6 | 20.4 | 15.7 | 12.2 | 16.3 | 14.3 | 14.3 | 31.3 | 24.5 | 60.4 | 38.7 | |
| Gemini-3.1-ProCategory=LinkS2Bench (tiny) Perf.2026.04 | 76.7 | 52.7 | 60 | 60.7 | 65.8 | 53.3 | 48.3 | 41.7 | 47.8 | 53.3 | 28.3 | 48.3 | 43.3 | 50 | 46.7 | 65 | 53.9 | |
| Qwen3.5-397B-A17BRank=1, Category=Open-source Models2026.04 | 66.7 | 45.6 | 73.3 | 36.1 | 58.7 | 31 | 50 | 39.5 | 40.2 | 23.9 | 37.5 | 41.2 | 34.2 | 26.1 | 55.1 | 66.7 | 49.3 | |
| RandomCategory=Baseline2026.04 | 50 | — | 25 | — | — | 25 | 25 | — | — | 12.5 | 20 | 25 | 19.2 | 16.7 | 33.3 | 50 | 33.3 |