Terminal-Bench Hard-AA

Terminal-Bench Hard (AA subset of Terminal-Bench 2.0): agentic terminal tasks; pass rate.

94 models · 153 results · data as of 2026-10-29 · artificialanalysis.ai ↗

More from this benchmark

3 published measures

Score vs any published measure3 measures: sub-scores, splits, counts

Highlight models

Click a point on the chart to highlight that exact effort; click it again to remove it.

54 more in settings

Chart settings

94 of 94 selected

darker is better within each column · click a column to sort

153 rows

#	Model	Effort	Terminal-Bench Hard-AA
1	GPT-5.6 Sol	max	65.9%
2	GPT-5.6 Sol	medium	62.9%
3	GPT-5.6 Terra	xhigh	62.9%
4	GPT-5.6 Sol	high	62.1%
5	GPT-5.6 Sol	xhigh	61.4%
6	GPT-5.5	xhigh	60.6%
7	GPT-5.6 Sol	low	60.6%
8	GPT-5.5	high	59.8%
9	Opus 4.8	max	58.3%
10	GPT-5.4	xhigh	57.6%
11	GPT-5.5	medium	57.6%
12	GPT-5.6 Terra	high	57.6%
13	GPT-5.6 Terra	max	57.6%
14	GPT-5.3 Codex	xhigh	53.0%
15	GPT-5.4 Mini	xhigh	52.3%
16	GPT-5.5	low	52.3%
17	Opus 4.7	max	51.5%
18	GLM 5.2	max	50.8%
19	Qwen 3.7 Max	max	50.8%
20	GPT-5.5	none	49.2%
21	GPT-5.2	xhigh	47.0%
22	Qwen 3.7 Plus		47.0%
23	Sonnet 4.6	none	46.2%
24	DeepSeek V4 Pro 0424	max	46.2%
25	Gemini 3.5 Flash	minimal	46.2%
26	Muse Spark		45.5%
27	GPT-5.1	high	45.5%
28	Kimi K2.7 Code		44.7%
29	GPT-5.6 Terra	low	43.9%
30	Kimi K2.6		43.9%