UReason: Benchmarking the Reasoning Paradox in Unified Multimodal Models

2026-02-09

AI Summary

This work investigates the practical impact of reasoning processes on image generation within unified multimodal models and identifies a “reasoning paradox”: although reasoning traces outperform direct generation, retaining intermediate reasoning steps degrades output quality, with distilled reasoning prompts yielding optimal results. This reveals contextual interference as a critical bottleneck. To systematically examine this phenomenon, the authors introduce UReason, a diagnostic benchmark comprising 2,000 samples across five fine-grained reasoning tasks, and employ chain-of-thought prompting, context-stripped generation, and a multimodal evaluation framework. Experiments across eight open-source unified models demonstrate the widespread presence of this paradox, indicating that current models are primarily constrained by contextual interference rather than insufficient reasoning capacity, thereby offering a new direction for effectively integrating reasoning and generation.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

To elicit capabilities for addressing complex and implicit visual requirements, recent unified multimodal models increasingly adopt chain-of-thought reasoning to guide image generation. However, the actual effect of reasoning on visual synthesis remains unclear. We present UReason, a diagnostic benchmark for reasoning-driven image generation that evaluates whether reasoning can be faithfully executed in pixels. UReason contains 2,000 instances across five task families: Code, Arithmetic, Spatial, Attribute, and Text reasoning. To isolate the role of reasoning traces, we introduce an evaluation framework comparing direct generation, reasoning-guided generation, and de-contextualized generation which conditions only on the refined prompt. Across eight open-source unified models, we observe a consistent Reasoning Paradox: Reasoning traces generally improve performance over direct generation, yet retaining intermediate thoughts as conditioning context often hinders visual synthesis, and conditioning only on the refined prompt yields substantial gains. Our analysis suggests that the bottleneck lies in contextual interference rather than insufficient reasoning capacity. UReason provides a principled testbed for studying reasoning in unified models and motivates future methods that effectively integrate reasoning for visual generation while mitigating interference.

Problem

• reasoning paradox
• multimodal models
• image generation
• chain-of-thought reasoning
• visual synthesis

Innovation

• reasoning paradox
• multimodal reasoning
• image generation
• contextual interference
• diagnostic benchmark