4D Spatio-temporal Reasoning on VLM4D
77.31Overall ScoreR4
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| R42025.12 | 77.31 | 78.17 | 77.33 | 77.61 | 77.42 | 76.13 | 76.4 | — | — | — | |
| SNOW2025.12 | 73.75 | 73.04 | 72.78 | 72.87 | 71.16 | 77.86 | 76.46 | — | — | — | |
| Gemini-2.5-Pro2025.12 | 62 | 64.6 | 62.9 | 63.5 | 54.8 | 80 | 57.3 | — | — | — | |
| Gemini-2.5-Pro2025.12 | 62 | 64.6 | 62.9 | 63.5 | 54.8 | 80 | 57.3 | — | — | — | |
| Gemini-2.5-ProModel Category=Proprietary Models2026.02 | 62 | — | — | — | — | — | — | — | 63.5 | 57.3 | |
| MLLM-4DModel Category=3D Spatial Reasoning Models, Base Model=Qwen3-VL-8B2026.02 | 61.08 | — | — | — | — | — | — | — | 63.111 | 54.42 | |
| GPT-4o2025.12 | 57.5 | 55.5 | 62.2 | 60 | 49.5 | 53.3 | 49.9 | — | — | — | |
| GPT-4o2025.12 | 57.5 | 55.5 | 62.2 | 60 | 49.5 | 53.3 | 49.9 | — | — | — | |
| GPT-4oModel Category=Proprietary Models2026.02 | 57.5 | — | — | — | — | — | — | — | 60 | 49.9 | |
| MLLM-4DModel Category=3D Spatial Reasoning Models, Base Model=Qwen2.5-VL-7B2026.02 | 57.013 | — | — | — | — | — | — | — | 59.416 | 49.74 | |
| Llama-4-Scout-17BParameters=17B2025.12 | 54.1 | 48.6 | 56.2 | 53.7 | 53.3 | 75.6 | 55.5 | — | — | — | |
| Llama-4-Scout-17B2025.12 | 54.1 | 48.6 | 56.2 | 53.7 | 53.3 | 75.6 | 55.5 | — | — | — | |
| Llama-4-Maverick-17BParameters=17B2025.12 | 53.6 | 52.6 | 54.3 | 53.8 | 53.3 | 51.1 | 53 | — | — | — | |
| Llama-4-Maverick-17B2025.12 | 53.6 | 52.6 | 54.3 | 53.8 | 53.3 | 51.1 | 53 | — | — | — | |
| Qwen2.5-VL-72BParameters=72B2025.12 | 53 | 54.3 | 52.5 | 53.1 | 49.5 | 80 | 52.6 | — | — | — | |
| Qwen2.5-VL-72B2025.12 | 53 | 54.3 | 52.5 | 53.1 | 49.5 | 80 | 52.6 | — | — | — | |
| Qwen2.5-VL-72BModel Category=Open-Source Models2026.02 | 53 | — | — | — | — | — | — | — | 53.1 | 52.6 | |
| Qwen3-VL-8BModel Category=Open-Source Models2026.02 | 52.2 | — | — | — | — | — | — | — | 52.1 | 52.4 | |
| Claude-Sonnet-42025.12 | 51.3 | 52.6 | 52.1 | 52.2 | 44 | 86.7 | 48.3 | — | — | — | |
| Claude-Sonnet-42025.12 | 51.3 | 52.6 | 52.1 | 52.2 | 44 | 86.7 | 48.3 | — | — | — | |
| InternVideo2.5-8BParameters=8B2025.12 | 50.7 | 57.2 | 50.5 | 52.7 | 44.3 | 46.7 | 44.5 | — | — | — | |
| InternVideo2.5-8B2025.12 | 50.7 | 57.2 | 50.5 | 52.7 | 44.3 | 46.7 | 44.5 | — | — | — | |
| InternVideo2.5-8BModel Category=Open-Source Models2026.02 | 50.7 | — | — | — | — | — | — | — | 52.7 | 44.5 | |
| VG-LLMModel Category=3D Spatial Reasoning Models, Base Model=Qwen2.5-VL-7B2026.02 | 46.52 | — | — | — | — | — | — | — | 49.56 | 37.38 | |
| Qwen2.5-VL-7BModel Category=Open-Source Models2026.02 | 43.8 | — | — | — | — | — | — | — | 43.3 | 45.6 | |
| LLaVA-NeXT-Video-7BModel Category=Open-Source Models2026.02 | 36.2 | — | — | — | — | — | — | — | 38.2 | 29.9 | |
| VLM-3RModel Category=3D Spatial Reasoning Models, Base Model=LLaVA-NeXT-Video-7B2026.02 | 33.92 | — | — | — | — | — | — | — | 36.9 | 24.75 | |
| Claude-Sonnet-4Access Type=Proprietary2026.03 | — | — | — | — | — | — | — | 52.6 | — | — | |
| EgoVLM GRPO (3B)*Parameters=3B, Access Type=Open-source MLLMs2026.03 | — | — | — | — | — | — | — | 46.8 | — | — | |
| Gemini-2.5-FlashMode=Zero-shot2025.12 | — | — | — | — | — | — | — | 54.7 | — | — | |
| Gemini2.5-ProAccess Type=Proprietary2026.03 | — | — | — | — | — | — | — | 64.6 | — | — | |
| GPT-4oAccess Type=Proprietary2026.03 | — | — | — | — | — | — | — | 55.5 | — | — | |
| Grok-2-VisionAccess Type=Proprietary2026.03 | — | — | — | — | — | — | — | 48.8 | — | — | |
| InternVideo2Parameters=8B, Access Type=Open-source MLLMs2026.03 | — | — | — | — | — | — | — | 35.6 | — | — | |
| Llava-OneVisionParameters=7B, Access Type=Open-source MLLMs2026.03 | — | — | — | — | — | — | — | 36.8 | — | — | |
| NVILA-Video-15BMode=Zero-shot2025.12 | — | — | — | — | — | — | — | 51.8 | — | — | |
| NVILA-Video-15BMode=Fine-tuned, Data=FoundationMotion2025.12 | — | — | — | — | — | — | — | 51.9 | — | — | |
| NVILA-Video-15BMode=Fine-tuned, Data=PLM 467k2025.12 | — | — | — | — | — | — | — | 52.9 | — | — | |
| NVILA-Video-8BMode=Zero-shot2025.12 | — | — | — | — | — | — | — | 49 | — | — | |
| NVILA-Video-8BMode=Fine-tuned, Data=FoundationMotion2025.12 | — | — | — | — | — | — | — | 52.4 | — | — | |
| NVILA-Video-8BMode=Fine-tuned, Data=PLM 467k2025.12 | — | — | — | — | — | — | — | 49.1 | — | — | |
| Qwen-2.5-VL-72BMode=Zero-shot2025.12 | — | — | — | — | — | — | — | 50.5 | — | — | |
| Qwen-2.5-VL-7BMode=Zero-shot2025.12 | — | — | — | — | — | — | — | 41.7 | — | — | |
| Qwen-2.5-VL-7BMode=Fine-tuned, Data=FoundationMotion2025.12 | — | — | — | — | — | — | — | 44.9 | — | — | |
| Qwen2.5-VLParameters=3B, Access Type=Open-source MLLMs2026.03 | — | — | — | — | — | — | — | 37.1 | — | — | |
| Qwen2.5-VLParameters=7B, Access Type=Open-source MLLMs2026.03 | — | — | — | — | — | — | — | 42.3 | — | — | |
| Qwen2.5-VL 3B + TGPOParameters=3B, Training Strategy=Temporal Global Policy Optimization (TGPO)2026.03 | — | — | — | — | — | — | — | 49.6 | — | — |