FinFIRST is a 123-row financial QA dataset on Hugging Face containing Chinese-language questions that require data retrieval and arithmetic on stocks, futures contracts, and macroeconomic indicators.

Each row includes a query, expected answer, scoring rubric, question decomposition, and structured metadata fields covering subject category, region, time constraint, and calculation type. Sample tasks involve computing stock purchase costs at minimum lot sizes, GDP year-over-year growth rates, and per-contract futures values with specified precision and unit requirements.