Terminal-Bench Hard

소프트웨어 공학, 시스템 관리, 데이터 처리 과제를 통해 터미널 환경에서 AI 역량을 평가하는 에이전트 벤치마크입니다.

모델: 147 현재, 432 전체
중앙값: 6.8%
범위: 0.0%–53.8%

점수 분포

가장 낮음: 0.0%
중앙값: 6.8%
최고: 53.8%

리더보드

지원 중단된 모델 포함

순위	모델	점수
1	Gemini 3.1 Pro Preview Google	53.8%
2	GPT-5.3 Codex (Xhigh) OpenAI	53.0%
3	KAT Coder Pro V2 KwaiKAT	49.2%
4	Qwen3.7 Plus Alibaba	47.0%
5	Kimi K2.7 Code Kimi	44.7%
6	MiniMax-M3 MiniMax	42.4%
7	Qwen3.5 397B A17B (Reasoning) Alibaba	40.9%
8	o3 OpenAI	37.1%
9	Gemma 4 31B (Reasoning) Google	36.4%
10	Nemotron 3 Ultra 550B A55B (Reasoning) NVIDIA	36.4%
11	Qwen3.5 397B A17B (Non-reasoning) Alibaba	35.6%
12	Qwen3.6 35B A3B (Reasoning) Alibaba	34.8%
13	Mistral Medium 3.5 Mistral	33.3%
14	North Mini Code Cohere	31.1%
15	Qwen3.5 122B A10B (Reasoning) Alibaba	31.1%
16	Gemma 4 31B (Non-reasoning) Google	30.3%
17	Qwen3.5 122B A10B (Non-reasoning) Alibaba	29.5%
18	JT-35B-Flash China Mobile	28.8%
19	Ring-2.6-1T InclusionAI	28.8%
20	Doubao Seed Code ByteDance Seed	26.5%
21	Qwen3.6 35B A3B (Non-reasoning) Alibaba	25.8%
22	Command A+ Cohere	25.0%
23	ERNIE 5.0 Thinking Preview Baidu	25.0%
24	Gemma 4 26B A4B (Non-reasoning) Google	25.0%
25	Nova 2.0 Pro Preview (Medium) Amazon	24.2%
26	Qwen3.5 9B (Reasoning) Alibaba	24.2%
27	gpt-oss-120b (High) OpenAI	23.5%
28	HyperNova 60B 2605 (High, Based on gpt-oss-120b) Multiverse Computing	23.5%
29	Trinity Large Thinking Arcee AI	22.7%
30	Nemotron Cascade 2 30B A3B NVIDIA	21.2%
31	Qwen3.5 Omni Plus Alibaba	21.2%
32	EXAONE 4.5 33B (Reasoning) LG AI Research	20.5%
33	Gemma 4 12B (Reasoning) Google	18.2%
34	JT-MINI China Mobile	18.2%
35	Qwen3.5 4B (Reasoning) Alibaba	18.2%
36	Qwen3.5 9B (Non-reasoning) Alibaba	18.2%
37	Qwen3 Coder Next Alibaba	18.2%
38	Mistral Small 4 (Reasoning) Mistral	17.4%
39	Nova 2.0 Lite (Medium) Amazon	17.4%
40	Nova 2.0 Pro Preview (Low) Amazon	17.4%
41	Cogito v2.1 (Reasoning) Deep Cogito	16.7%
42	Nova 2.0 Lite (High) Amazon	16.7%
43	Nova 2.0 Pro Preview (Non-reasoning) Amazon	16.7%
44	Apriel-v1.6-15B-Thinker ServiceNow	14.4%
45	Gemma 4 26B A4B (Reasoning) Google	13.6%
46	NVIDIA Nemotron 3 Nano 30B A3B (Reasoning) NVIDIA	13.6%
47	Magistral Medium 1.2 Mistral	12.9%
48	HyperCLOVA X SEED Think (32B) Naver	12.1%
49	NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning) NVIDIA	12.1%
50	Gemma 4 12B (Non-reasoning) Google	11.4%

1
Gemini 3.1 Pro Preview
Google
53.8%
2
GPT-5.3 Codex (Xhigh)

데이터 출처: Artificial Analysis

Command Palette

Terminal-Bench Hard

점수 분포

리더보드