OmniCap-IF: leaderboard
OmniCap-IF (2026) benchmarks instruction following in captioning videos with sound: each instruction stacks format and content constraints, checked by rule-based tools and by a gpt-5-mini judge answering questions against the caption. The headline is the constraint satisfaction rate (%) over all constraints; higher is better.
Metric: Overall constraint satisfaction rate (CSR, %): share of all constraints satisfied, on multi-constraint captioning instructions on videos with sound; format constraints are checked by rule-based tools on LLM-extracted structure, content constraints by temporal-IoU tools or a gpt-5-mini judge answering binary and multiple-choice questions against the caption; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 80.65 | #94 |
| 2 | Gemini 3 Flash | 79.5 | #164 |
| 3 | MiMo-V2.5 | 76.22 | #354 |
| 4 | Qwen3 Omni 30B A3B Instruct | 62.65 | #904 |
| 5 | Qwen2.5-Omni-7B | 49.19 | #1789 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omnicap-if · How It Works · Data refreshed daily, snapshot 2026-11-13.