SIS-Bench - Action Recognition: leaderboard
Metric: Accuracy (%; 686 action recognition 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 2030. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 66.5 |
| 2 | GPT-5.4 | 57.3 |
| 3 | InternVL3.5-8B | 56.1 |
| 4 | Qwen 3 VL 8B Instruct | 55.1 |
| 5 | Qwen 3 VL 4B Instruct | 53.4 |
| 6 | Qwen 3 VL 8B (Thinking) | 53.4 |
| 7 | Qwen 3.5 Plus | 52.8 |
| 8 | Kimi K2.5 | 50.9 |
| 9 | Step3 VL 10B | 50 |
| 10 | GLM-4.1V-9B (Thinking) | 48.1 |
| 11 | Seed 1.8 | 43.7 |
| 12 | Qwen 2.5 VL 7B Instruct | 43.3 |
| 13 | InternVL3-14B | 42.7 |
| 14 | Qwen 3 VL 30B A3B Instruct | 39.7 |
| 15 | Qwen 2 VL 7B Instruct | 38.6 |
Interactive version: theaggregate.ai/benchmark?slug=sis-bench-action-recognition · How It Works · Data refreshed daily, snapshot 2026-09-29.