# harbor-framework/terminal-bench-1

A benchmark for LLMs on complicated tasks in the terminal

Repository: https://github.com/harbor-framework/terminal-bench-1
Canonical: https://ross.abutalabs.com/products/terminal-bench-1
Homepage: https://www.tbench.ai
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-11T05:19:08+00:00

## Health v2 (maintenance only)
Score: 62/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 92, release rhythm 35, longevity 42
- inputs: {"age_days": 593, "days_push": 53, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2554, forks 568 (observed 2026-08-28T04:07:00.680973+00:00)

## What it is
Terminal-Bench is a benchmark and execution harness for evaluating LLM agents on complex, real-world tasks in a terminal sandbox. It combines a dataset of ~100+ terminal tasks with tooling that connects language models to the environment for reproducible evaluation.

## Use cases
- benchmark llm agents on terminal tasks
- evaluate how well a model can use a shell autonomously
- compare coding agents on end-to-end system tasks
- run reproducible agent evaluations in a sandbox
- contribute new challenging tasks for agent testing

## When to choose
- you need a standardized, reproducible benchmark for terminal-based agent capabilities
- you are building or stress-testing LLM agents that operate in a shell
- you want to compare models on real-world system-level tasks like compiling code or configuring servers

## When to avoid
- you need a general-purpose chat or reasoning benchmark rather than terminal tasks
- you want a lightweight unit-testing tool rather than an agent evaluation harness
- your agents operate in GUI or web environments rather than text terminals

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, agent-framework, testing, cli
- domain: large-language-models, developer-tools
- platform: python, cli
- tags: llm-benchmark, terminal-agents, evaluation-harness, agent-evaluation, sandbox, ai-agents, command-line, docker, linux, macos

## Member repositories
- harbor-framework/terminal-bench-1 (main) score 62

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:00.680973+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:23:44.562020+00:00, confidence not recorded.
  - readme: https://github.com/harbor-framework/terminal-bench-1 (fetched 2026-08-28T04:07:00.680973+00:00, sha c4bb406dbba0)
  - homepage: https://www.tbench.ai (fetched 2026-08-29T10:06:22.893422+00:00, sha e42bdd88110e)
- Data as of 2026-08-30T08:39:29.467469+00:00.