Video-IFBench: leaderboard
Video-IFBench (2026) evaluates instruction following of multimodal LLMs on videos, checking each response against a constraint checklist by an LLM judge and programmatic verifiers. Task-gated instruction satisfaction rate over all four instruction types, in percent; higher is better.
Metric: Task-gated instruction satisfaction rate (%; TISR over all four instruction types; 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 54.5 | #135 |
| 2 | Gemini 3 Flash | 49.5 | #159 |
| 3 | Qwen 3.5 397B A17B (Thinking) | 46.1 | #242 (Qwen 3.5 397B A17B) |
| 4 | Qwen 3.5 122B A10B (Thinking) | 41.7 | #329 (Qwen 3.5 122B A10B) |
| 5 | Doubao-Seed-2.0-Pro-260215 | 40.8 | #149 |
| 6 | Qwen 3.5 27B (Thinking) | 37.5 | #358 (Qwen 3.5 27B) |
| 7 | Qwen 3.5 35B A3B (Thinking) | 37.2 | #480 (Qwen 3.5 35B A3B) |
| 8 | Gemma 4 31B (IT) | 35.4 | #343 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 33.5 | #451 |
| 10 | Qwen 3.5 9B (Thinking) | 32.2 | #803 (Qwen 3.5 9B) |
| 11 | Qwen 3.5 397B A17B (Non-reasoning) | 30.4 | #242 (Qwen 3.5 397B A17B) |
| 12 | GPT-5.4 | 30 | #121 |
| 13 | Gemma 4 26B A4B (IT) | 29.7 | #545 |
| 14 | Qwen 3 VL 235B A22B Instruct | 25.8 | #521 |
| 15 | Qwen 3.5 122B A10B (Non-reasoning) | 25.3 | #329 (Qwen 3.5 122B A10B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=video-ifbench · How It Works · Data refreshed daily, snapshot 2026-11-12.