LLM Stats (MVBench): leaderboard
Video understanding across 20 temporal tasks including action recognition, event ordering, velocity estimation, and moving direction identification.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 Flash | 77.8 | #65 |
| 2 | Qwen 3.5 122B A10B | 76.6 | #173 |
| 3 | Qwen 3.6 27B | 75.5 | #170 |
| 4 | Qwen 3.5 35B A3B | 74.8 | #256 |
| 5 | Qwen 3.5 27B | 74.6 | #187 |
| 6 | Qwen 3.6 35B A3B | 74.6 | #226 |
| 7 | Qwen 3 VL 32B (Thinking) | 73.2 | #294 (Qwen 3 VL 32B) |
| 8 | Qwen 3 VL 32B Instruct | 72.8 | #283 |
| 9 | Qwen 3 VL 30B A3B Instruct | 72.3 | #384 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 72 | #351 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 2.5 VL 72B Instruct | 70.4 | #377 |
| 12 | Qwen2.5-Omni-7B | 70.3 | #624 |
| 13 | Qwen 2.5 VL 7B Instruct | 69.6 | #654 |
| 14 | Qwen 3 VL 4B (Thinking) | 69.3 | #480 (Qwen 3 VL 4B) |
| 15 | Qwen 3 VL 8B (Thinking) | 69 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-mvbench · How It Works · Data refreshed daily, snapshot 2026-10-20.