SIS-Bench - Action Recall: leaderboard
Metric: Accuracy (%; 804 action recall questions; four-option multiple choice on UAV videos (at most 32 frames), zero-shot, open models through vLLM with at most 128 generated tokens and proprietary models through their APIs; higher is better). Source: arxiv.org. Saturation forecast: Around June 2028. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Seed 1.8 | 59.2 |
| 2 | Kimi K2.5 | 53 |
| 3 | GPT-5.4 | 49.9 |
| 4 | Gemini 3 Flash (Preview) | 42.4 |
| 5 | Qwen 3.5 Plus | 42.4 |
| 6 | GLM-4.1V-9B (Thinking) | 34.8 |
| 7 | Qwen 3 VL 8B (Thinking) | 33.6 |
| 8 | Qwen 2 VL 7B Instruct | 33.3 |
| 9 | Qwen 3 VL 8B Instruct | 32.2 |
| 10 | InternVL3.5-8B | 31.8 |
| 11 | Qwen 2.5 VL 7B Instruct | 29.2 |
| 12 | Qwen 3 VL 4B Instruct | 29.1 |
| 13 | Step3 VL 10B | 28.6 |
| 14 | Qwen 3 VL 30B A3B Instruct | 28.4 |
| 15 | InternVL3-14B | 24.9 |
Interactive version: theaggregate.ai/benchmark?slug=sis-bench-action-recall · How It Works · Data refreshed daily, snapshot 2026-09-29.