{
  "id": "terminal-bench",
  "code": "PTL-0203",
  "term": "Terminal-Bench",
  "aliases": [],
  "category": "benchmarks",
  "definition": "Terminal-Bench is a benchmark of hard, realistic tasks that an AI agent must complete by running commands in a containerized Linux terminal, graded by automated tests of the final state of the environment.",
  "description": "Grading checks outcomes, not the commands used, so agents can solve tasks any way that works. It measures the command-line skills that coding and DevOps agents rely on, such as building software, configuring systems, and debugging.",
  "example": null,
  "broader": [
    "benchmark"
  ],
  "narrower": [],
  "related": [
    "coding-agent",
    "swe-bench",
    "code-execution-tool"
  ],
  "introduced": null,
  "added": "2026-10-10",
  "sources": [
    {
      "title": "Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces",
      "authors": "Merrill et al.",
      "year": 2026,
      "url": "https://arxiv.org/abs/2601.11868"
    }
  ],
  "url": "https://protologue.com/t/terminal-bench/",
  "citation": "Protologue. (2026). Terminal-Bench. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0203). https://protologue.com/t/terminal-bench/"
}