GameplayQA - Intent Identification: leaderboard
Metric: Accuracy (%) on the 148 Intent Identification questions (the model must identify the intent or goal behind actions) of GameplayQA (multiple-choice questions on synchronized first-person 3D gameplay videos with structured distractors, zero-shot; Gemini models take the whole video, the others 1 frame per second up to 32 frames); higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 64.9 | #401 |
| 2 | Qwen 3 VL 235B A22B Instruct | 63.5 | #264 |
| 3 | Seed 1.6 Flash | 63.1 | #427 |
| 4 | Gemini 3 Flash | 62.8 | #93 |
| 5 | Claude Sonnet 4.5 | 61.5 | #138 |
| 6 | Gemini 2.5 Flash | 60.8 | #237 |
| 7 | Seed-1.6 | 60.1 | #257 |
| 8 | Gemini 2.5 Pro | 59.5 | #145 |
| 9 | Qwen 3 VL 30B A3B Instruct | 56.8 | #365 |
| 10 | GPT-5 | 54.1 | #91 |
| 11 | Gemma 3 27B (IT) | 50.7 | #509 |
| 12 | Gemma 3 12B (IT) | 50 | #655 |
| 13 | GPT-5 Mini | 50 | #176 |
| 14 | Gemma 3 4B (IT) | 46.6 | #971 |
| 15 | Claude Haiku 4.5 | 46.6 | #271 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=gameplayqa-intent-identification · How It Works · Data refreshed daily, snapshot 2026-10-11.