Video-IFBench - Multi: leaderboard

Metric: Task-gated instruction satisfaction rate (%; TISR on multi-task instructions; checklist-based instruction following on 1.5K video samples with 39 semantic and format constraint types; a response counts only if it addresses the active task (the correct branch for conditional instructions) and satisfies every checklist item, judged by Qwen3.5-397B-A17B-Instruct or verified programmatically; Gemini and Doubao get video at 1 fps, GPT-5.4 50 frames, vision-only models frames plus timestamped subtitles). Source: arxiv.org. Saturation forecast: Around February 2027. 38 models tracked.

Top models

#ModelScore
1Gemini 3 Pro58.8
2Gemini 3 Flash56.4
3Qwen 3.5 397B A17B (Thinking)52.8
4Doubao-Seed-2.0-Pro-26021551.1
5Qwen 3.5 122B A10B (Thinking)50.1
6Qwen 3.5 35B A3B (Thinking)49.1
7Qwen 3.5 27B (Thinking)46.3
8Qwen 3 VL 235B A22B (Thinking)45
9Gemma 4 31B (IT)44.4
10GPT-5.443.2
11Gemma 4 26B A4B (IT)41.1
12Qwen 3.5 397B A17B (Non-reasoning)40.8
13Qwen 3.5 9B (Thinking)39.8
14Qwen 3.5 122B A10B (Non-reasoning)36.8
15Qwen 3 VL 235B A22B Instruct35.1

Interactive version: theaggregate.ai/benchmark?slug=video-ifbench-multi · How It Works · Data refreshed daily, snapshot 2026-09-26.