[Paper Review] UniFinEval: Towards Unified Evaluation of Financial Multimodal Models across Text, Images and Videos

Zhi Yang, Lingfeng Zeng+17 more|arXiv (Cornell University)|Jan 9, 2026

Stock Market Forecasting Methods0 citations

TL;DR

UniFinEval is a manually constructed, bilingual (Chinese-English) multimodal benchmark evaluating financial MLLMs across text, images, and videos in five core scenarios with cross-modal multi-hop reasoning. It compares 10 mainstream models under Zero-Shot and Zero-Shot CoT settings, highlighting remaining gaps to financial experts.

ABSTRACT

Multimodal large language models are playing an increasingly significant role in empowering the financial domain, however, the challenges they face, such as multimodal and high-density information and cross-modal multi-hop reasoning, go beyond the evaluation scope of existing multimodal benchmarks. To address this gap, we propose UniFinEval, the first unified multimodal benchmark designed for high-information-density financial environments, covering text, images, and videos. UniFinEval systematically constructs five core financial scenarios grounded in real-world financial systems: Financial Statement Auditing, Company Fundamental Reasoning, Industry Trend Insights, Financial Risk Sensing, and Asset Allocation Analysis. We manually construct a high-quality dataset consisting of 3,767 question-answer pairs in both chinese and english and systematically evaluate 10 mainstream MLLMs under Zero-Shot and CoT settings. Results show that Gemini-3-pro-preview achieves the best overall performance, yet still exhibits a substantial gap compared to financial experts. Further error analysis reveals systematic deficiencies in current models. UniFinEval aims to provide a systematic assessment of MLLMs' capabilities in fine-grained, high-information-density financial environments, thereby enhancing the robustness of MLLMs applications in real-world financial scenarios. Data and code are available at https://github.com/aifinlab/UniFinEval.

Motivation & Objective

• Assess the capability boundaries of multimodal large language models (MLLMs) in high-information-density financial environments.
• Provide a unified, cross-modality benchmark aligned with real-world financial workflows.
• Enable evaluation of cross-modal consistency and multi-hop reasoning in finance.
• Identify common failure modes to guide robust financial AI deployment.

Proposed method

• Manual construction of a 3,767-question bilingual (Chinese-English) dataset.
• Five financial scenarios: Financial Statement Auditing, Company Fundamental Reasoning, Industry Trend Insights, Financial Risk Sensing, Asset Allocation Analysis.
• Support for full modality inputs: text, images, and videos with cross-modal combinations (text-image, text-video, image-video, text-image-video).
• Two evaluation settings: Zero-Shot and Zero-Shot CoT, with output extraction standardized by Qwen-Max for robust judging.
• Expert-driven quality control with four-stage validation to ensure alignment with real-world financial logic.

Figure 1: UniFinEval is manually constructed and supports full-modality inputs including text, images, and videos. It is equipped with cross-modal reasoning capabilities and features high information density while closely aligning with real financial business practices.

Experimental results

Research questions

• RQ1 Can current MLLMs perform integrated cross-modal reasoning in high-information-density financial tasks?
• RQ2 How close are existing models to financial expert performance across perception, reasoning, and decision-making tasks?
• RQ3 What are the dominant error modes when processing multimodal financial information?
• RQ4 How does Chain-of-Thought prompting impact performance in finance-specific cross-modal tasks?
• RQ5 What are the limitations of current benchmarks in simulating real-world financial decision loops?

Key findings

• Gemini-3-pro-preview achieves the best overall Zero-Shot performance with an average of 73.8%.
• Most models improve under CoT but the gains are limited across tasks.
• Humans (experts) greatly outperform all models, with substantial gaps in ITI and AAA scenarios.
• Error analysis shows major issues in image perception and cross-modal alignment, with notable numerical computation weaknesses.
• Models struggle with cross-modal multi-hop reasoning and maintaining long-term logical consistency in high-density tasks.

Figure 2: UniFinEval covers five major financial scenarios and constructs datasets spanning text, images, videos, as well as multiple cross-modal combinations. It features high-information-density and manually construct data, together with dedicated designs for cross-modal consistency checking and m