Long Audio-Video Understanding Benchmark for Omnimodal LLMs

LVOmniBench is an evaluation benchmark for omnimodal large language models focused on long-form audio-video understanding. It was created by KD-TAO and launched in March 2026.

Use Cases

• Evaluate OmniLLM performance on long-form audio-video understanding tasks.
• Benchmark model capabilities on the AVQA (Audio-Visual Question Answering) task.
• Assess multimodal reasoning across the Lvomni Bench and Regionus tags.

Strengths

• Benchmark launched in March 2026, indicating recent development.
• Focuses on the specific domain of long audio-video understanding.

Limitations

• Dataset size, row count, and column structure are unknown.
• Sample data and file formats are unavailable for inspection.
• Potential for limited scope or task variety based on available tags.

Provenance

Source

KD-TAO via Hugging Face.

Freshness

Last updated on 2026-03-19.