---
format: "aidr-story-markdown/v1"
id: "b68f09c8fe91f3b134f1fa6cdc790c90581490904a982bb2cc4ec53472e3f344"
canonical_url: "https://aidr.today/b68f09c8?lang=en"
title: "OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-16T00:00:00.000Z"
category: "Data"
topics: ["llm","data-engineering"]
source_urls: ["https://huggingface.co/papers/2610.18829"]
summary: "OmniCompass is a holistic benchmark for omni-modal video grounding and understanding, featuring two evaluation tracks: OmniCompass-VL for vision-language models and OmniCompass-AV for omni-modal models. Each track includes temporal grounding, spatial grounding, and question answering tasks, with OmniCompass-VL containing 820 videos and 2,460 instances, and OmniCompass-AV containing 240 videos and 720 instances."
---

# OmniCompass: A Holistic Benchmark for Omni\-Modal Video Grounding and Understanding

> [Open the canonical story](<https://aidr.today/b68f09c8?lang=en>)

**Published:** 2026-10-16T00:00:00.000Z
**Category:** Data
**Topics:** llm, data\-engineering

## Summary

OmniCompass is a holistic benchmark for omni\-modal video grounding and understanding, featuring two evaluation tracks: OmniCompass\-VL for vision\-language models and OmniCompass\-AV for omni\-modal models\. Each track includes temporal grounding, spatial grounding, and question answering tasks, with OmniCompass\-VL containing 820 videos and 2,460 instances, and OmniCompass\-AV containing 240 videos and 720 instances\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.18829>)