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

Hugging Face Daily Papers Papers

Summary

X-Stream introduces the first benchmark for multi-stream video understanding, evaluating MLLMs as multiplexers across multiple concurrent streams. The study reveals that current MLLMs achieve only about 50% accuracy, exposing significant limitations in handling multiple streams.

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.
Original Article
View Cached Full Text

Cached at: 06/02/26, 03:24 AM

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

Source: https://huggingface.co/papers/2606.02482 Authors:

,

,

,

,

,

,

,

,

,

,

Abstract

X-Stream introduces the first benchmark for multi-stream streaming understanding, revealing significant limitations of current MLLMs in handling concurrent streams.

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 acrossmulti-window,multi-view, andmulti-device scenarios. Crucially, our dataset is constructed using a noveldual-verification pipelinethat prevents over-reliance on a single stream. Furthermore, we pioneer the conceptualization ofmulti-modal large language models(MLLMs) as naive multiplexers, systematically evaluating their performance through the lens ofSignal Multiplexing Theory. Our extensive online inference experiments reveal a stark reality: state-of-the-art MLLMs struggle significantly withconcurrent 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.

View arXiv pageView PDFProject pageGitHub17Add to collection

Get this paper in your agent:

hf papers read 2606\.02482

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.02482 in a model README.md to link it from this page.

Datasets citing this paper1

#### spw2000/X-stream Updated25 minutes ago • 669 • 2

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.02482 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

RefereeBench: Are Video MLLMs Ready to be Multi-Sport Referees

arXiv cs.CL

RefereeBench introduces the first large-scale benchmark with 925 curated sports videos and 6,475 QA pairs to evaluate whether video MLLMs can reliably act as multi-sport referees. Evaluation of state-of-the-art models shows current MLLMs fall short (≤60% accuracy), struggling with rule application and temporal grounding despite their generic video understanding capabilities.

MUX: Continuous Reasoning via Multiplexed Tokens

arXiv cs.AI

MUX proposes a method for lossless continuous reasoning by distilling discrete reasoning steps into multiplexed latent tokens that encode a superposition of subwords, achieving higher bandwidth and enabling parallel exploration in language model reasoning tasks.