ITG NowHome › AI glossary › Terminal-Bench
What is Terminal-Bench?
Terminal-Bench is a test that checks whether an AI agent can finish real tasks by typing commands in a terminal.
- Category Benchmarks
- Updated 20 Oct 2026
Terminal-Bench, explained simply
Terminal-Bench gives an AI agent a task to do on a computer using only the command line. That is the text window developers use.
Tasks can include setting up software or fixing a broken setup. The result is checked automatically.
This matters because it tests doing the work, not just answering questions. Higher is better.
The score depends on both the model and the agent tool around it, so compare like with like.
On ITG Now, the AI models page lists Terminal-Bench results for agent skill, with a link to the source.
Related terms
BenchmarkTerminalCLIAI coding agentAI agentSWE-bench
Where you'll see it on ITG Now
- AI newsDeepSeek says free model update tops a coding test
- AI modelGLM-5.3
- AI modelClaude Fable 5.1
- AI modelClaude Opus 5.5
- AI modelGPT-6 Astra
- AI modelGPT-6 Luna
Questions people ask about Terminal-Bench
Is a higher Terminal-Bench score better?
Yes. It means the agent finished more tasks correctly.
What is a terminal?
It is a text window where you type commands to control a computer. Many developer tools use it.
Terminal-Bench vs SWE-bench: what is the difference?
SWE-bench fixes bugs in code projects. Terminal-Bench tests tasks done through the command line.