Dataset · birdsql
Updated BIRD Mini-Dev benchmark for Text-to-SQL agents
birdsql/bird_mini_dev
500 verified questions and reference SQL queries for SQLite, MySQL, and PostgreSQL simultaneously.
What's inside
The 2026 update of the popular benchmark contains 500 aligned questions in English for three DBMSs: SQLite, MySQL, and PostgreSQL. Annotations, evidence prompts, and reference SQL queries underwent manual re-verification following community feedback to eliminate ambiguities and errors.
What you can do with it
- Evaluating Text-to-SQL agents capable of exploring database schemas and writing queries
- Comparing SQL generation across different dialects (SQLite, MySQL, PostgreSQL)
- Validating enterprise analytical assistants for working with databases
- Tasks
- Table question answering, Question answering
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 5.7K
- Likes
- 15
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("birdsql/bird_mini_dev")