Terminal-Bench Hard

Un benchmark agentique qui évalue les capacités de l'IA dans des environnements de terminal via des tâches d'ingénierie logicielle, d'administration système et de traitement de données.

Modèles: 147 actuel, 432 au total
Médiane: 6.8%
Plage: 0.0%–53.8%

Distribution des scores

Le plus bas: 0.0%
Médiane: 6.8%
Le plus élevé: 53.8%

Classement

Inclure les modèles obsolètes

Rang	Modèle	Score
101	Llama 3.1 Nemotron Ultra 253B v1 (Reasoning) NVIDIA	2.3%
102	Mi:dm K 2.5 Pro Korea Telecom	2.3%
103	Sarvam 30B (High) Sarvam	2.3%
104	Solar Open 100B (Reasoning) Upstage	2.3%
105	EXAONE 4.0 32B (Non-reasoning) LG AI Research	1.5%
106	Granite 4.0 Micro IBM	1.5%
107	Llama 4 Scout Meta	1.5%
108	Nova Micro Amazon	1.5%
109	NVIDIA Nemotron Nano 9B V2 (Reasoning) NVIDIA	1.5%
110	Qwen3 Omni 30B A3B Instruct Alibaba	1.5%
111	Sarvam 105B (High) Sarvam	1.5%
112	Command A Cohere	0.8%
113	Jamba Reasoning 3B AI21 Labs	0.8%
114	Llama 3.2 Instruct 11B (Vision) Meta	0.8%
115	NVIDIA Nemotron Nano 9B V2 (Non-reasoning) NVIDIA	0.8%
116	Olmo 3 7B Think Allen Institute for AI	0.8%
117	Tri-21B-Think Trillion Labs	0.8%
118	Apertus 70B Instruct Swiss AI Initiative	0.0%
119	Apertus 8B Instruct Swiss AI Initiative	0.0%
120	Exaone 4.0 1.2B (Non-reasoning) LG AI Research	0.0%
121	Exaone 4.0 1.2B (Reasoning) LG AI Research	0.0%
122	Gemma 3 270M Google	0.0%
123	Granite 4.0 350M IBM	0.0%
124	Granite 4.0 H 350M IBM	0.0%
125	Granite 4.0 H 1B IBM	0.0%
126	Hermes 4 - Llama-3.1 70B (Non-reasoning) Nous Research	0.0%
127	Jamba 1.7 Mini AI21 Labs	0.0%
128	LFM2 24B A2B Liquid AI	0.0%
129	LFM2.5-1.2B-Instruct Liquid AI	0.0%
130	LFM2.5-1.2B-Thinking Liquid AI	0.0%
131	LFM2.5-VL-1.6B Liquid AI	0.0%
132	MiniCPM-V 4.6 1.3B OpenBMB	0.0%
133	MiniCPM5-1B (Reasoning) OpenBMB	0.0%
134	MiniCPM5-1B (Non-reasoning) OpenBMB	0.0%
135	Ministral 3 3B Mistral	0.0%
136	Molmo 7B-D Allen Institute for AI	0.0%
137	Molmo2-8B Allen Institute for AI	0.0%
138	Nanbeige4.1-3B Nanbeige	0.0%
139	NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning) NVIDIA	0.0%
140	Olmo 3.1 32B Instruct Allen Institute for AI	0.0%
141	Olmo 3.1 32B Think Allen Institute for AI	0.0%
142	Olmo 3 7B Instruct Allen Institute for AI	0.0%
143	Phi-4 Mini Instruct Microsoft	0.0%
144	Qwen3.5 0.8B (Reasoning) Alibaba	0.0%
145	Qwen3.5 0.8B (Non-reasoning) Alibaba	0.0%
146	Reka Flash 3 Reka AI	0.0%
147	Tiny Aya Global Cohere	0.0%

101
Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)
NVIDIA
2.3%
102
Mi:dm K 2.5 Pro

Données de Artificial Analysis

Command Palette

Terminal-Bench Hard

Distribution des scores

Classement