VCIFBench: leaderboard
VCIFBench evaluates complex instruction following in video understanding over 306 constraint-rich instructions spanning content, format, style and structure requirements, judged by a hybrid rule-based, executable and GPT-5 pipeline. This board reports Instruction pass rate (%): the share of the 306 instructions on which every annotated constraint is jointly satisfied; higher is better.
Metric: Instruction pass rate (%): the share of the 306 instructions on which every annotated constraint is jointly satisfied; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 52.94 | #137 |
| 2 | GPT-5.4 | 51.96 | #123 |
| 3 | Gemini 3 Flash | 51.63 | #164 |
| 4 | GLM-4.1V-9B (Thinking) | 26.8 | #1060 (GLM-4.1V-9B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vcifbench · How It Works · Data refreshed daily, snapshot 2026-11-13.