Multi-Turn Multimodal Financial Reasoning Benchmark

FinMTM is a multi-turn multimodal benchmark for financial reasoning and agent evaluation. It contains image and text data, with over 10,000 instances indicated by its 'Size Categories1 Kn10 K' tag. The dataset was created by researchers from HiThink Research, Wuhan University, Zhejiang University, Nanyang Technological University, and Shanghai Institute of Technology.

Use Cases

• Evaluate financial reasoning agents on multi-turn tasks using multimodal text and image data.
• Benchmark the performance of large language models on complex financial question-answering scenarios.
• Train models for financial document analysis that require interpreting both textual and visual information.

Strengths

• Contains over 10,000 instances, providing a substantial testbed for evaluation.
• Features multimodal data, combining both text and image modalities for complex reasoning.
• Designed specifically for multi-turn interactions, simulating realistic financial agent scenarios.

Limitations

• Specific row counts, column definitions, and sample data are not publicly available for review.
• The dataset's primary language is Chinese ('Languagezh'), which may limit applicability for English-only models.
• The lack of accessible column details makes it difficult to assess the specific data structure and features.

Provenance

Source

HiThink-Research on Hugging Face.

Freshness

Last updated on 2026-01-28, indicating recent maintenance.

Geography

Region tag indicates 'us', but primary language is Chinese, suggesting potential geographic focus.