Egocentric Video Temporal Reasoning on EgoTempo
45.2EgoTempo ScoreQwen2.5-VL 3B + TGPO
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL 3B + TGPOBackbone=Qwen2.5-VL, Model Scale=3B, Optimization Method=TGPO2026.03 | 45.2 | |
| EgoVLM GRPO (3B)*Model Scale=3B, Optimization Method=GRPO2026.03 | 40.8 | |
| GPT-4oModel Type=Proprietary2026.03 | 40.1 | |
| Gemini-FlashModel Type=Proprietary2026.03 | 39.1 | |
| Qwen2-VLModel Scale=72B, Model Type=Open-source MLLMs2026.03 | 28.4 | |
| LLaVA-OneVisionModel Scale=72B, Model Type=Open-source MLLMs2026.03 | 26.5 | |
| Qwen2-VLModel Scale=7B, Model Type=Open-source MLLMs2026.03 | 26.1 | |
| LLaVA-OneVisionModel Scale=7B, Model Type=Open-source MLLMs2026.03 | 23.3 | |
| LLaVA-NeXT-VideoModel Scale=34B, Model Type=Open-source MLLMs2026.03 | 15.7 | |
| Claude-3.5-SonnetModel Type=Proprietary2026.03 | 13.1 |