VRBench: leaderboard
Long narrative video benchmark for multi-step reasoning, assessing both final outcomes and process-level reasoning chains over temporally grounded questions.
Metric: Overall (imported). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 68.68 | #343 |
| 2 | Claude 3.7 Sonnet | 68.15 | #249 |
| 3 | InternVL2.5-78B | 62.31 | #431 |
| 4 | Gemini 2.0 Flash (Thinking) | 60.97 | #342 (Gemini 2.0 Flash) |
| 5 | O1 Preview | 60.14 | #193 |
| 6 | DeepSeek R1 | 57.13 | #251 |
| 7 | Qwen 2.5 VL 7B | 56.52 | #643 |
| 8 | DeepSeek V3 | 56.06 | #321 |
| 9 | Qwen 2 VL 7B | 54.08 | #750 |
| 10 | Qwen 2.5 72B Instruct | 53.51 | #454 |
| 11 | QwQ-32B | 52.52 | #421 |
| 12 | Llama 3.3 70B Instruct | 49.84 | #534 |
| 13 | Qwen 2.5 7B Instruct | 48.29 | #862 |
| 14 | InternLM3-8B-Instruct | 47.81 | #867 |
| 15 | QwQ 32B-Preview | 35.9 | #651 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vrbench · How It Works · Data refreshed daily, snapshot 2026-10-20.