ConvFinQA

ConvFinQA: Evaluates financial analysis, accounting, market reasoning, and quantitative business tasks.

9rows
execution_accuracyprimary metric
2026-05-27sampled

Metadata

ID
convfinqa
Category
Finance
Release
2022-10-07

Metrics

Execution accuracy, Program accuracy

Latest Results

Rows are parsed from the ConvFinQA paper arXiv LaTeX neural-symbolic main results table.

Rank Subject Execution accuracy Model Match Provenance Sampled
1 Human Expert Performance 89.44 — Imported 2026-05-27
2 FinQANet-Gold (RoBERTa-large) 77.32 — Imported 2026-05-27
3 FinQANet (RoBERTa-large) 68.9 — Imported 2026-05-27
4 FinQANet (RoBERTa-base) 64.95 — Imported 2026-05-27
5 FinQANet (BERT-large) 61.14 — Imported 2026-05-27
6 T-5(large) 58.66 — Imported 2026-05-27
7 GPT-2(medium) 58.19 — Imported 2026-05-27
8 FinQANet (BERT-base) 55.03 — Imported 2026-05-27
9 General Crowd Performance 46.9 — Imported 2026-05-27