Terminal-Bench 4.0

按 Terminal-Bench (Stanford, Laude Institute) 發布的原樣呈現，指標和模型都以它為準。點模型名檢視本站詳情，點箭頭回到原榜單。

發布方

Terminal-Bench (Stanford, Laude Institute)

授權條款

未註明

更新頻率

持續更新

本站抓取時間

2026-10-07

條目數

35

1. Claude Opus 5.5 64.8%
2. Claude Sonnet 5.5 61.8%
3. GPT 6 Astra 58.2%
4. GPT 6.1 Sol 58.2%
5. Claude Fable 5.1 57.9%
6. Claude Opus 5 53.9%
7. GPT 6 Sol 49.4%
8. Claude Fable 5 44.5%
9. GLM 5.3 41.8%
10. Grok 4.7 37.6%
11. GPT 5.6 Sol 37.3%
12. GLM 5.3 Flash 35.8%

篩選模型排序依據Terminal-Bench 4.0Cost per trial只看已識別的模型

35 個有分數

#	模型	Terminal-Bench 4.0 ↑	Cost per trial ↓
1	Claude Opus 5.5Claude Code · maxAnthropic	64.8%61.7–68.0	$14.30
2	Claude Sonnet 5.5Claude Code · maxAnthropic	61.8%58.9–64.8	$22.22
3	GPT 6 AstraCodex · maxOpenAI	58.2%55.4–61.0	$9.90
4	GPT 6.1 SolCodex · maxOpenAI	58.2%55.1–61.3	$1.92
5	Claude Fable 5.1Claude Code · maxAnthropic	57.9%54.1–61.6	$18.92
6	Claude Fable 5.1Claude Code · xhighAnthropic	57.9%54.5–61.2	$14.76
7	GPT 6 AstraCodex · highOpenAI	57.9%54.9–60.8	$6.88
8	GPT 6 AstraCodex · xhighOpenAI	57.9%55.2–60.6	$7.12
9	Claude Fable 5.1Claude Code · highAnthropic	54.5%51.1–58.0	$12.08
10	GPT 6 AstraCodex · mediumOpenAI	54.2%51.6–56.9	$5.80
11	Claude Fable 5.1Claude Code · mediumAnthropic	53.9%50.5–57.3	$8.58
12	Claude Opus 5Claude Code · xhighAnthropic	53.9%50.8–57.1	$18.44
13	Claude Opus 5Claude Code · maxAnthropic	51.8%48.4–55.2	$18.09
14	GPT 6 AstraCodex · lowOpenAI	50.6%47.9–53.4	$4.72
15	Claude Opus 5Claude Code · highAnthropic	50.3%46.6–54.0	$14.13
16	GPT 6 SolCodex · maxOpenAI	49.4%46.2–52.6	$4.40
17	Claude Opus 5Claude Code · mediumAnthropic	44.9%41.0–48.7	$9.67
18	Claude Fable 5Claude Code · maxAnthropic	44.5%40.7–48.4	$22.02
19	Claude Fable 5.1Claude Code · lowAnthropic	43.3%39.7–46.9	$7.15
20	GLM 5.3Claude Code · maxZ AI	41.8%38.6–45.1	$8.27
21	Grok 4.7Grok Build · xhighSpaceXAI	37.6%34.0–41.1	$11.16
22	GPT 5.6 SolCodex · maxOpenAI	37.3%33.5–41.1	$7.70
23	GLM 5.3 FlashClaude Code · offZ AI	35.8%32.2–39.3	$8.74
24	Claude Opus 5Claude Code · lowAnthropic	34.9%30.9–38.8	$7.25
25	Qwen 3.8 MaxClaude Code · 0902 · maxAlibaba	27.0%23.2–30.8	$14.05
26	Claude Opus 4.8Claude Code · maxAnthropic	23.6%20.1–27.2	$19.64
27	GPT 5.6 TerraCodex · maxOpenAI	21.5%18.3–24.8	$5.25
28	Grok 4.6Grok Build · highSpaceXAI	20.3%17.2–23.4	$10.88
29	Gemini 3.8 Flashmini-SWE-agent · highGoogle	19.1%15.7–22.4	$5.54
30	GPT 5.6 LunaCodex · maxOpenAI	17.3%14.4–20.1	$1.05
31	GPT 6 LunaCodex · maxOpenAI	16.4%13.6–19.1	$0.22
32	Muse Spark 1.3Muse Code · xhighMeta	14.6%11.6–17.5	—
33	Grok 4.5Grok Build · highSpaceXAI	12.4%9.8–15.0	$6.35
34	Claude Sonnet 5Claude Code · maxAnthropic	12.4%9.4–15.5	$29.10
35	Gemini 3.7 Flashmini-SWE-agent · highGoogle	11.2%8.8–13.7	$3.82

每一列測的是什麼

Terminal-Bench 4.0

越高越好 · 代理式程式設計

Cost per trial

越低越好