Dataset · InternScience
AI agent benchmark for autonomous scientific research
InternScience/ResearchClawBench
Benchmark tasks for testing the ability of coding agents to independently conduct scientific research from data analysis to paper writing.
What's inside
The dataset is a benchmark with fewer than 1,000 records in tabular and text formats. In the tasks, models receive raw data and must independently conduct research and prepare a publication, the results of which are then strictly compared against real human scientific papers.
What you can do with it
- Testing autonomous AI agents and R&D process automation systems
- Comparing analytical report generation quality and scientific conclusion reproduction across different LLMs
- Tasks
- Other
- Size
- up to 1K rows
- Downloads this month
- 19.7K
- Likes
- 16
How to load it
from datasets import load_dataset
ds = load_dataset("InternScience/ResearchClawBench")