Video Reasoning on LongVideoBench
59LongVideoBench ScoreTriage
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TriageModel=LLaVA-Video-7B, Retention Ratio=100%2026.01 | 59 | — | |
| TriageModel=LLaVA-OneVision-7B, Retention Ratio=25%2026.01 | 58.6 | — | |
| TriageModel=LLaVA-OneVision-7B, Retention Ratio=100%2026.01 | 58.3 | — | |
| TriageModel=LLaVA-Video-7B, Retention Ratio=75%2026.01 | 58.3 | — | |
| TriageModel=LLaVA-Video-7B, Retention Ratio=50%2026.01 | 58 | — | |
| TriageModel=LLaVA-OneVision-7B, Retention Ratio=75%2026.01 | 57.9 | — | |
| TriageModel=LLaVA-OneVision-7B, Retention Ratio=50%2026.01 | 57.6 | — | |
| PyramidDropModel=LLaVA-OneVision-7B, Retention Ratio=50%2026.01 | 57.1 | — | |
| TriageModel=Qwen2-VL-7B, Retention Ratio=75%2026.01 | 57 | — | |
| VanillaModel=LLaVA-Video-7B, Retention Ratio=100%2026.01 | 56.8 | — | |
| VanillaModel=LLaVA-OneVision-7B, Retention Ratio=100%2026.01 | 56.7 | — | |
| TriageModel=Qwen2-VL-7B, Retention Ratio=100%2026.01 | 56.7 | — | |
| FastVModel=LLaVA-OneVision-7B, Retention Ratio=50%2026.01 | 56.4 | — | |
| DyCokeModel=LLaVA-OneVision-7B, Retention Ratio=50%2026.01 | 56.3 | — | |
| PyramidDropModel=LLaVA-Video-7B, Retention Ratio=50%2026.01 | 56.2 | — | |
| DyCokeModel=LLaVA-Video-7B, Retention Ratio=50%2026.01 | 56.1 | — | |
| TriageModel=Qwen2-VL-7B, Retention Ratio=50%2026.01 | 55.9 | — | |
| VanillaModel=Qwen2-VL-7B, Retention Ratio=100%2026.01 | 55.7 | — | |
| TriageModel=LLaVA-Video-7B, Retention Ratio=25%2026.01 | 55 | — | |
| TriageModel=Qwen2-VL-7B, Retention Ratio=25%2026.01 | 54.8 | — | |
| PyramidDropModel=Qwen2-VL-7B, Retention Ratio=50%2026.01 | 54.6 | — | |
| FastVModel=LLaVA-Video-7B, Retention Ratio=50%2026.01 | 54 | — | |
| FastVModel=Qwen2-VL-7B, Retention Ratio=50%2026.01 | 52.8 | — | |
| DyCokeModel=Qwen2-VL-7B, Retention Ratio=50%2026.01 | 52.7 | — | |
| AdaFocusReasoning Mode=AutoThink2026.05 | — | 57.85 | |
| FFRDistillation Method=FFR, Teacher Model=Qwen3-32B2026.04 | — | 52.6 | |
| FFRDistillation Method=FFR, Teacher Model=Qwen3-235B2026.04 | — | 54.2 | |
| FFR (GLM-4.5V)Distillation Method=FFR, Teacher Model=GLM-4.5V2026.04 | — | 55.3 | |
| ProcessThinker (outcome + process)Backbone=QWEN3-VL-8B, Training strategy=GRPO, Reward type=outcome + process2026.04 | — | 74.6 | |
| ProcessThinker (outcome-only)Backbone=QWEN3-VL-8B, Training strategy=GRPO, Reward type=outcome-only2026.04 | — | 74.2 | |
| ProcessThinker (process-only)Backbone=QWEN3-VL-8B, Training strategy=GRPO, Reward type=process-only2026.04 | — | 75.4 | |
| PROCESSTHINKER-SFTBackbone=QWEN3-VL-8B, Training strategy=SFT2026.04 | — | 68.5 | |
| Qwen2.5-VL-7B (Base)Reasoning Mode=✗2026.05 | — | 60.9 | |
| QWEN3-VL-8B-INSTRUCTBackbone=QWEN3-VL-8B2026.04 | — | 71.5 | |
| SFTDistillation Method=SFT, Teacher Model=Qwen3-32B2026.04 | — | 50.4 | |
| SFTDistillation Method=SFT, Teacher Model=Qwen3-235B2026.04 | — | 54.5 | |
| VIDEO-R1-7BBackbone=QWEN2.5-VL2026.04 | — | 58.3 | |
| Video-R1-SFTDistillation Method=SFT, Teacher Model=Baseline2026.04 | — | 47.6 | |
| VideoAuto-R1Reasoning Mode=AutoThink2026.05 | — | 56.5 | |
| VideoChat-R1.5Reasoning Mode=Think-Only2026.05 | — | 61.4 |