Fin-Eva

蚂蚁集团、上海财经大学联合推出金融评测集Fin-Eva Version 1.0，覆盖财富管理、保险、投资研究等多个金融场景以及金融专业主题学科，总评测题数目达到1.3w+。
蚂蚁数据源包括各业务领域数据、互联网公开数据，经过数据脱敏、文本聚类、语料精筛、数据改写等处理过程后，结合金融领域专家的评审构建而成。 上海财经大学数据源主要基于相关领域权威性考试的各类真题和模拟题对知识大纲的要求，由上海财经大学统计与管理学院张立文副教授课题组牵头，金融学院闵敏副教授及其他各学院老师协助完成，所有数据均为原创，这保证了数据源的准确性和权威性。
蚂蚁部分涵盖金融认知、金融知识、金融逻辑、内容生成以及安全合规五大类能力33个子维度共8445个测评题； 上财部分涵盖金融，经济，会计和证书等四大领域，包括4661个问题，涵盖34个不同的学科。
Fin-Eva Version 1.0 全部采用单选题这类有固定答案的问题，配合相应指令让模型输出标准格式。
目前Fin-Eva Version 1.0对外开放评测数据，未来持续迭代并开放平台化评测托管服务，为行业提供一站式评估。

Fin-Eva Version 1.0, a financial evaluation benchmark jointly launched by Ant Group and Shanghai University of Finance and Economics (SUFE), covers multiple financial scenarios such as wealth management, insurance, investment research, as well as specialized financial subject areas, with a total of over 13,000 evaluation questions.
Ant Group's data sources include data from various business domains and publicly available internet data. It is constructed after undergoing processes such as data anonymization, text clustering, corpus fine screening, and data rewriting, combined with reviews from financial domain experts.
SUFE's data sources are mainly based on the knowledge outline requirements of various real questions and simulation questions from authoritative examinations in related fields. It was led by the research group of Associate Professor Zhang Liwen from the School of Statistics and Management of SUFE, assisted by Associate Professor Min Min from the School of Finance and teachers from other colleges. All data are original, which ensures the accuracy and authority of the data sources.
The Ant Group section covers 8,445 evaluation questions under 33 sub-dimensions across five capability categories: financial cognition, financial knowledge, financial logic, content generation, and security compliance. The SUFE section covers 4,661 questions across four major fields: finance, economics, accounting, and certification, involving 34 different disciplines.
Fin-Eva Version 1.0 exclusively uses single-choice questions with fixed answers, paired with corresponding instructions to require models to output in standard formats.
Currently, the evaluation data of Fin-Eva Version 1.0 is open to the public. In the future, it will continue to iterate and launch platform-based evaluation hosting services to provide one-stop assessment for the industry.