Dataset · harborframework
Regularly updated Terminal-Bench benchmark
harborframework/terminal-bench
A compact, versioned multimodal benchmark for testing models on terminal tasks.
What's inside
A mirror of the actively maintained Terminal-Bench benchmark hosted on GitHub. The dataset contains fewer than 1,000 examples and includes texts, images, and documents. New versions are published periodically as tags in a single repository, with the main branch always pointing to the latest release.
What you can do with it
- Evaluating AI agent and LLM performance in command line and terminal environments
- Comparing different model versions on a single regularly updated benchmark
- Testing multimodal agents on tasks featuring documents and screenshots
- Size
- up to 1K rows
- Downloads this month
- 205.7K
- Likes
- 26
How to load it
from datasets import load_dataset
ds = load_dataset("harborframework/terminal-bench")