---
license: apache-2.0
task_categories:
- text-generation
tags:
- benchmark
- agentic
- self-improvement
---

# AutoDataBench

AutoDataBench is a benchmark to evaluate an agent’s ability to autonomously write new tasks that feed the self-improvement loop. Given an original benchmark task and a record of the target model attempting it, the agent must produce a new task for the same suite that meets practical acceptance criteria on validity, novelty, difficulty, and behavioural coverage.

- **Paper**: [AutoDataBench: Can Agents Write the Data That Feeds the Self-Improvement Loop?](https://huggingface.co/papers/2609.35025)
- **Project page**: [https://autodatabench.com/](https://autodatabench.com/)
- **Code**: [https://github.com/StarDewXXX/AutoDataBench](https://github.com/StarDewXXX/AutoDataBench)

The dataset contains rollouts and rubrics for three executable agent benchmarks: `automationbench`, `terminal-bench`, and `tb-science`. It is designed to measure whether individual produced tasks meet the standards of a production data pipeline.

For usage and detailed scoring methodology, please refer to the [GitHub repository](https://github.com/StarDewXXX/AutoDataBench).