X-Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding

2026-06-01

AI Summary

Existing video understanding benchmarks are confined to a single-stream paradigm, making them inadequate for evaluating models’ ability to handle concurrent multi-stream inputs and perform cross-stream reasoning in real-world scenarios. To address this gap, this work introduces X-Stream, the first benchmark specifically designed for multi-stream streaming comprehension, encompassing multi-window, multi-view, and multi-device settings, along with a dual-validation data construction pipeline. For the first time, signal multiplexing theory is leveraged to systematically evaluate multimodal large language models (MLLMs) as multiplexers. Experimental results reveal that state-of-the-art MLLMs achieve only around 50% performance under concurrent stream processing and exhibit weak active reasoning capabilities, thereby exposing fundamental limitations in current multiplexing mechanisms.

Abstract

While video streaming understanding has made significant strides, real-world applications, such as live sports broadcasting, autonomous driving, and multi-screen collaboration, inherently demand continuous, multi-stream interactions. However, existing benchmarks are confined to single-stream paradigms, leaving a critical gap in evaluating online, cross-stream reasoning. To bridge this, we introduce X-Stream, the first benchmark dedicated to multi-stream streaming understanding. Comprising 4,220 rigorously curated QA pairs across 932 videos, X-Stream evaluates 11 subtasks across multi-window, multi-view, and multi-device scenarios. Crucially, our dataset is constructed using a novel dual-verification pipeline that prevents over-reliance on a single stream. Furthermore, we pioneer the conceptualization of multi-modal large language models (MLLMs) as naive multiplexers, systematically evaluating their performance through the lens of Signal Multiplexing Theory. Our extensive online inference experiments reveal a stark reality: state-of-the-art MLLMs struggle significantly with concurrent streams, achieving only about 50% score and exhibiting poor proactive ability. Ultimately, X-Stream exposes the trade-off of current multiplexing schemes, providing both a practical evaluation protocol and empirical guidance for next-generation multi-stream agents.

Problem

• multi-stream understanding
• video streaming
• cross-stream reasoning
• online inference
• multi-modal large language models

Innovation

• multi-stream understanding
• MLLMs as multiplexers
• Signal Multiplexing Theory
• dual-verification pipeline
• online cross-stream reasoning