Video-IFBench - Single: leaderboard

Metric: Task-gated instruction satisfaction rate (%; TISR on single-task instructions with 1 to 15 constraints; 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 August 2027. 38 models tracked.

Top models

#ModelScore
1Gemini 3 Pro52.3
2Gemini 3 Flash49.2
3Qwen 3.5 397B A17B (Thinking)48.5
4Qwen 3.5 122B A10B (Thinking)45.6
5Doubao-Seed-2.0-Pro-26021544.6
6Gemma 4 31B (IT)43
7Qwen 3.5 27B (Thinking)40.1
8Qwen 3.5 35B A3B (Thinking)37.9
9Qwen 3 VL 235B A22B (Thinking)37.7
10Qwen 3.5 397B A17B (Non-reasoning)36.5
11Qwen 3.5 9B (Thinking)35.9
12Gemma 4 26B A4B (IT)35.7
13GPT-5.434.7
14Qwen 3 VL 235B A22B Instruct31.9
15Qwen 3.5 27B (Non-reasoning)29.1

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