Terminal-Bench
Harbor nativeTerminal-Bench is a benchmark for measuring agents' abilities to complete tasks using a terminal.
- Samples
- 66
- Released
- 2025
- Scores
- 145
How to run
ltb · terminal-bench-2
$ npm i -g lettokenburn-cli$ ltb bench run terminal-bench/terminal-bench-2@1 --benchmark terminal-bench-2 -a claude-code -m anthropic/claude-opus-4-8$ ltb bench push
Harbor-native: you can also call Harbor directly.
harbor
$ harbor run -d terminal-bench/terminal-bench-2@1 -a claude-code -m anthropic/claude-opus-4-8
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
12 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.4 | 75.1 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Claude Opus 4.6 | 65.4 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Claude Sonnet 4.6 | 59.1 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Gemini 3.1 Pro | 56.2 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| GLM-5 | 56.2 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Qwen 3.5 | 52.5 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Grok 3 | 52.0 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| Kimi K2.5 | 50.8 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| MiniMax M2.5 | 42.2 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| DeepSeek V3.2 | 39.6 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| MiMo-V2-Flash | 38.5 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
| GPT-oss 120B | 18.7 | N/A | Self-reported | low | Jun 1, 2026 | tbench.ai |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
Terminal-Bench 2.0 (Terminal Bench) — 命令行/DevOps操作. Scoring: 准确率. Difficulty: 极高. Imported from benchmark_research.md (2026-06).