Terminal-Bench 4.0
2026-10-06 · 终端 Agent
- 测什么
- 终端里的 Agent 任务:安装环境、改配置、跑命令、把自动化做完。
- 怎么测
- 公开榜单 Terminal-Bench 4.0 的全部 display 行。列与来源一致:Agent、模型、档位、通过率、发布日期、Agent 机构、模型机构、Tokens、成本。并列名次保留来源名次,不改成连续序号。
- 分数含义
- 通过率用榜单上的 display accuracy,例如 64.8% ± 3.1%,不用未展示的原始小数。成本用 display cost。Grok 4.7 的成本格写着 partial: 324/330。
- 注意
- 它不衡量界面好不好看。同一通过率下,跑分成本可以差一个数量级。
- 更新
- 榜单 updated_at 为 2026-10-06。本页于 2026-10-10 核对全部 35 行。
完整榜单
35
| # | Agent | 模型 | 档位 | 通过率 | 发布日期 | Agent 机构 | 模型机构 | Tokens | 成本 |
|---|
| 1 | Claude Code | Opus 5.5 | max | 64.8% ± 3.1% | Sep 22, 2026 | Anthropic | Anthropic | 8.0B | $4.7k |
| 2 | Claude Code | Sonnet 5.5 | max | 61.8% ± 2.9% | Sep 28, 2026 | Anthropic | Anthropic | 19.4B | $7.3k |
| 3 | Codex | GPT-6 Astra | max | 58.2% ± 2.8% | Sep 3, 2026 | OpenAI | OpenAI | 1.5B | $3.3k |
| 3 | Codex | GPT-6.1 Sol | max | 58.2% ± 3.1% | Sep 29, 2026 | OpenAI | OpenAI | 1.5B | $634.24 |
| 5 | Claude Code | Fable 5.1 | max | 57.9% ± 3.8% | Sep 1, 2026 | Anthropic | Anthropic | 2.7B | $6.2k |
| 5 | Codex | GPT-6 Astra | xhigh | 57.9% ± 2.7% | Sep 3, 2026 | OpenAI | OpenAI | 1.2B | $2.4k |
| 5 | Codex | GPT-6 Astra | high | 57.9% ± 3.0% | Sep 3, 2026 | OpenAI | OpenAI | 1.2B | $2.3k |
| 5 | Claude Code | Fable 5.1 | xhigh | 57.9% ± 3.4% | Sep 1, 2026 | Anthropic | Anthropic | 2.3B | $4.9k |
| 9 | Claude Code | Fable 5.1 | high | 54.5% ± 3.4% | Sep 1, 2026 | Anthropic | Anthropic | 2.2B | $4.0k |
| 10 | Codex | GPT-6 Astra | medium | 54.2% ± 2.7% | Sep 3, 2026 | OpenAI | OpenAI | 1.1B | $1.9k |
| 11 | Claude Code | Fable 5.1 | medium | 53.9% ± 3.4% | Sep 1, 2026 | Anthropic | Anthropic | 1.6B | $2.8k |
| 11 | Claude Code | Opus 5 | xhigh | 53.9% ± 3.2% | Jul 24, 2026 | Anthropic | Anthropic | 6.9B | $6.1k |
| 13 | Claude Code | Opus 5 | max | 51.8% ± 3.4% | Jul 24, 2026 | Anthropic | Anthropic | 6.5B | $6.0k |
| 14 | Codex | GPT-6 Astra | low | 50.6% ± 2.8% | Sep 3, 2026 | OpenAI | OpenAI | 889.8M | $1.6k |
| 15 | Claude Code | Opus 5 | high | 50.3% ± 3.7% | Jul 24, 2026 | Anthropic | Anthropic | 5.5B | $4.7k |
| 16 | Codex | GPT-6 Sol | max | 49.4% ± 3.2% | Sep 22, 2026 | OpenAI | OpenAI | 5.0B | $1.5k |
| 17 | Claude Code | Opus 5 | medium | 44.8% ± 3.8% | Jul 24, 2026 | Anthropic | Anthropic | 3.7B | $3.2k |
| 18 | Claude Code | Fable 5 | max | 44.5% ± 3.8% | Jun 9, 2026 | Anthropic | Anthropic | 3.8B | $7.3k |
| 19 | Claude Code | Fable 5.1 | low | 43.3% ± 3.6% | Sep 1, 2026 | Anthropic | Anthropic | 1.3B | $2.4k |
| 20 | Claude Code | GLM-5.3 | max | 41.8% ± 3.2% | Aug 14, 2026 | Anthropic | Z.ai | 8.7B | $2.7k |