Video-MME-v2 - Level 3 Complex Reasoning (w/o subs): leaderboard
Video-MME-v2 group score without subtitles on level 3, complex reasoning across time. Video-MME-v2 groups four interrelated questions and scores each group non-linearly, so credit depends on consistency and a complete reasoning chain, not only per-question accuracy. 0-100, higher is better; same run per model as the headline board.
Metric: Level 3 Score w/o sub (%). Source: video-mme-v2.netlify.app. Saturation forecast: Around March 2028. 49 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 30.15 | #82 |
| 2 | Doubao-Seed-2.0-Pro-260215 | 26.21 | #86 |
| 3 | Gemini 3 Flash | 24.95 | #103 |
| 4 | Kimi K2.5 | 24.8 | #152 |
| 5 | GPT-5 | 21.33 | #105 |
| 6 | Gemma 4 31B | 15.98 | #212 |
| 7 | Qwen 3.5 Omni Plus | 14.53 | #111 |
| 8 | Qwen 3 VL 235B A22B (Thinking) | 14.49 | #236 |
| 9 | Qwen 3 VL 235B A22B Instruct | 13.89 | #280 |
| 10 | Qwen 2.5 VL 72B Instruct | 12.61 | #383 |
| 11 | Qwen 3 VL 8B (Thinking) | 11.11 | #423 |
| 12 | Qwen 3 VL 8B Instruct | 10.72 | #419 |
| 13 | Qwen 3 VL 4B Instruct | 10.26 | #510 |
| 14 | Qwen 3 VL 30B A3B (Thinking) | 10.02 | #348 |
| 15 | Qwen 3 VL 4B (Thinking) | 9.76 | #548 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=video-mme-v2-level-3-complex-reasoning-w-o-subs · How It Works · Data refreshed daily, snapshot 2026-11-05.