OmniBench: Towards The Future of Universal Omni-Language Models

TL;DR

OmniBench introduces a comprehensive benchmark for evaluating models' ability to process and reason across visual, acoustic, and textual modalities simultaneously, revealing current limitations and guiding future research in tri-modal AI systems.

Contribution

The paper presents OmniBench, a new benchmark with high-quality annotations for tri-modal understanding, and curates OmniInstruct, an instruction tuning dataset to improve omni-language models.

Findings

Open-source OLMs show limited reasoning in tri-modal tasks.

Most models perform below 50% accuracy on tri-modal benchmarks.

Existing training paradigms often overlook multi-modal context construction.

Abstract

Recent advancements in multimodal large language models (MLLMs) have aimed to integrate and interpret data across diverse modalities. However, the capacity of these models to concurrently process and reason about multiple modalities remains underexplored, partly due to the lack of comprehensive modality-wise benchmarks. We introduce OmniBench, a novel benchmark designed to rigorously evaluate models' ability to recognize, interpret, and reason across visual, acoustic, and textual inputs simultaneously. We define language models capable of such tri-modal processing as the omni-language models (OLMs). OmniBench is distinguished by high-quality human annotations, ensuring that accurate responses require integrated understanding and reasoning across all three modalities. Our main findings reveal that: i) open-source OLMs exhibit critical limitations in instruction-following and reasoning capabilities within tri-modal contexts; and ii) most baselines models perform poorly (below 50% accuracy) even when provided with alternative textual representations of images or/and audio. These results suggest that the ability to construct a consistent context from text, image, and audio is often overlooked in existing MLLM training paradigms. To address this gap, we curate an instruction tuning dataset of 84.5K training samples, OmniInstruct, for training OLMs to adapt to tri-modal contexts. We advocate for future research to focus on developing more robust tri-modal integration techniques and training strategies to enhance OLMs. Codes, data and live leaderboard could be found at https://m-a-p.ai/OmniBench.