Skip to content

Dataset · IntelligenceLab

Long-Horizon Terminal Tasks Benchmark

IntelligenceLab/Long-Horizon-Terminal-Bench

LHTB evaluates how LLM agents maintain useful work in a containerized terminal over hundreds of steps.

What's inside

Long-Horizon Terminal-Bench is a benchmark for LLM agents operating in a containerized terminal for hundreds of steps. LHTB 2.0 has 77 tasks and LHTB 1.0 has 46, counting soft solves with reward >= 0.95. Data is presented in document, tabular, and text formats in English, with a size under 1K.

What you can do with it

  • Evaluating agentic LLMs on long terminal tasks
  • Comparing containerized agent systems using LHTB
  • Checking agent robustness against multi-step failures
Tasks
Text generation
Languages
English
Size
up to 1K rows
Downloads this month
24.3K
Likes
144

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("IntelligenceLab/Long-Horizon-Terminal-Bench")