AEGIS: Exploring the Limit of World Knowledge Capabilities for Unified Multimodal Models

[📂 GitHub] [🆕 Blog] [📜 Paper]

Summary

The capability of Unified Multimodal Models (UMMs) to apply world knowledge across diverse tasks remains a critical, unresolved challenge. Existing benchmarks fall short, offering only siloed, single-task evaluations with limited diagnostic power. To bridge this gap, we propose AEGIS (i.e., Assessing Editing, Generation, Interpretation-Understanding for Super-intelligence), a comprehensive multi-task benchmark covering visual understanding, generation, editing, and interleaved generation. AEGIS comprises 1,050 challenging, manually-annotated questions spanning 21 topics (including STEM, humanities, daily life, etc.) and 6 reasoning types. To concretely evaluate the performance of UMMs in world knowledge scope without ambiguous metrics, we further propose Deterministic Checklist-based Evaluation (DCE), a protocol that replaces ambiguous prompt-based scoring with atomic “Y/N” judgments, to enhance evaluation reliability. Our extensive experiments reveal that most UMMs exhibit severe world knowledge deficits and that performance degrades significantly with complex reasoning. Additionally, simple plug-in reasoning modules can partially mitigate these vulnerabilities, highlighting a promising direction for future research. These results highlight the importance of world-knowledge-based reasoning as a critical frontier for UMMs.

Contribution

The main contributions of this work are as follows:

• Comprehensive Multi-Task Benchmark: Assesses Visual Understanding, Generation, Editing, and Interleaved Generation simultaneously.
• Extensive Knowledge Coverage: 1,050 questions across 21 topics (STEM, Humanities, Daily Life) and 6 reasoning types.
• Deterministic Evaluation (DCE): A novel checklist-based protocol that replaces ambiguous scores with atomic "Yes/No" judgments for reliability.
• In-depth Diagnosis: Reveals severe world knowledge deficits in SOTA UMMs and the impact of reasoning complexity.

Data Statistics

AEGIS covers three general domains (i.e., STEM, humanities, and daily life) with 21 diverse topics. Each topic contains 15 prompts for visual understanding, generation, and editing, as well as 5 visual interleaved generation questions to measure complex generative capabilities. Furthermore, AEGIS incorporates six distinct reasoning types into the majority of its prompts, requiring UMMs to possess inherent reasoning capabilities to complete each request.

Visualization

Usage

Please refer to our GitHub repository for usage details.

Citation

If you find this work helpful in your research, please consider citing:

@misc{aegis,
      title={AEGIS: Exploring the Limit of World Knowledge Capabilities for Unified Mulitmodal Models}, 
      author={Jintao Lin, Bowen Dong, Weikang Shi, Chenyang Lei, Suiyun Zhang, Rui Liu, Xihui Liu},
      year={2026},
      eprint={2601.00561},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2601.00561}, 
}