Long-horizon video understanding on MM-Lifelong (train@month)
82.5AccuracyHuman
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Human2026.05 | 82.5 | 31.2 | |
| ReMAMethod Category=Agentic Method, Backbone=GPT-52026.05 | 17.62 | 9.91 | |
| VAMMethod Category=Agentic Method2026.05 | 17.11 | 3.65 | |
| ReMAMethod Category=Agentic Method, Backbone=Qwen3VL-A22B2026.05 | 14.23 | 6.01 | |
| GPT-5Method Category=End-to-End MLLM2026.05 | 10.15 | 1.39 | |
| Qwen3-VL-235B-A22BMethod Category=End-to-End MLLM2026.05 | 9.09 | 0.39 | |
| Qwen3-VL-30B-A3BMethod Category=End-to-End MLLM2026.05 | 8.33 | 0.48 | |
| Nemotron-v2-12BMethod Category=End-to-End MLLM2026.05 | 7.52 | 0.19 | |
| Video-XL-2-8BMethod Category=End-to-End MLLM2026.05 | 6.02 | 0 | |
| LongVT-7BMethod Category=Agentic Method2026.05 | 5.83 | 1.71 | |
| Eagle-2.5-8BMethod Category=End-to-End MLLM2026.05 | 3.76 | 1.59 |