Agent-Omni: Test-Time Multimodal Reasoning via Model Coordination for Understanding Anything

TL;DR

Agent-Omni is a flexible, modular framework that coordinates existing foundation models for robust, multi-modal reasoning across text, images, audio, and video without retraining.

Contribution

It introduces a master-agent system that enables cross-modal reasoning by coordinating pre-existing models, avoiding costly fine-tuning and supporting diverse modalities.

Findings

Achieves state-of-the-art performance on multi-modal benchmarks.

Demonstrates effective cross-modal reasoning without retraining.

Ensures transparency and interpretability in multi-modal tasks.

Abstract

Multimodal large language models (MLLMs) have shown strong capabilities but remain limited to fixed modality pairs and require costly fine-tuning with large aligned datasets. Building fully omni-capable models that can integrate text, images, audio, and video remains impractical and lacks robust reasoning support. In this paper, we propose an Agent-Omni framework that coordinates existing foundation models through a master-agent system, enabling flexible multimodal reasoning without retraining. The master agent interprets user intent, delegates subtasks to modality-specific agents, and integrates their outputs into coherent responses. Extensive experiments across text, image, audio, video, and omni benchmarks show that Agent-Omni consistently achieves state-of-the-art performance, particularly on tasks requiring complex cross-modal reasoning. Its agent-based design enables seamless integration of specialized foundation models, ensuring adaptability to diverse inputs while maintaining transparency and interpretability. In addition, the framework is modular and easily extensible, allowing future improvements as stronger models become available.