Terminal-Bench-Science 0.1

Un benchmark de 70 tâches portant sur des workflows de recherche, rédigées et relues par des experts de domaine en sciences de la vie, physique, sciences de la Terre, mathématiques et ingénierie, chacune réalisée dans un terminal et vérifiée par son propre ensemble de tests.

Poids de l'Indice d'intelligence: 15%
Modèles: 109 actuel, 135 au total
Médiane: 7.6%
Plage: 0.0%–73.3%

Distribution des scores

Le plus bas: 0.0%
Médiane: 7.6%
Le plus élevé: 73.3%

Classement

Inclure les modèles obsolètes

Rang	Modèle	Score	Coût / tâche	Temps / tâche	Tokens de sortie / tâche
1	Claude Fable 5.2 (Max, Default Fallback) Anthropic	73.3%	$45.21	73m 49s	455K
2	Claude Fable 5.2 (High, Default Fallback) Anthropic	70.0%	$24.28	43m 19s	242K
3	Claude Fable 5.2 (Xhigh, Default Fallback) Anthropic	67.6%	$31.37	54m 42s	339K
4	Gemini 4 Carbon (High) Google	64.8%	$19.97	55m 59s	276K
5	GPT-6 Astra (Max) OpenAI	63.3%	$17.48	71m 32s	161K
6	Gemini 4 Carbon (Medium) Google	62.4%	$17.57	56m 19s	249K
7	Claude Opus 5.5 (Xhigh, Default Fallback) Anthropic	61.9%	$14.12	34m 26s	224K
8	Gemini 4 Carbon (Low) Google	61.9%	$18.89	64m 33s	250K
9	Claude Fable 5.2 (Medium, Default Fallback) Anthropic	60.5%	$10.98	21m 47s	119K
10	GPT-6 Astra (High) OpenAI	59.5%	$8.66	28m 31s	67.3K
11	Claude Opus 5.5 (Max, Default Fallback) Anthropic	59.0%	$19.97	43m 47s	307K
12	GPT-6.1 Sol (Max) OpenAI	58.1%	$3.34	58m 5s	196K
13	GPT-6.1 Terra (Max) OpenAI	57.6%	$2.87	31m 22s	138K
14	GPT-6 Astra (Xhigh) OpenAI	54.8%	$7.91	27m 27s	64.1K
15	Claude Fable 5.2 (Low, Default Fallback) Anthropic	53.8%	$8.78	17m 45s	98.9K
16	Claude Sonnet 5.5 (Max, Default Fallback) Anthropic	53.3%	$23.76	55m 13s	509K
17	GPT-6 Astra (Medium) OpenAI	53.3%	$7.20	20m 9s	51.6K
18	Gemini 4 Argon (High) Google	52.9%	$24.39	71m 30s	291K
19	Claude Sonnet 5.5 (Xhigh, Default Fallback) Anthropic	52.4%	$12.77	37m 2s	309K
20	GPT-6.1 Sol (Xhigh) OpenAI	50.5%	$2.30	23m 37s	79K
21	Claude Opus 5.5 (High, Default Fallback) Anthropic	49.0%	$10.02	24m 29s	154K
22	GPT-6.1 Sol (High) OpenAI	46.2%	$1.61	15m 43s	53.7K
23	Claude Opus 5.5 (Medium, Default Fallback) Anthropic	43.3%	$6.80	17m 44s	110K
24	GPT-6.1 Terra (Xhigh) OpenAI	41.9%	$1.04	14m 6s	60.8K
25	GPT-6.1 Sol (Medium) OpenAI	37.1%	$0.87	8m 22s	29.6K
26	GPT-6 Astra (Low) OpenAI	37.1%	$4.12	12m 4s	28.7K
27	Muse Wonder Full (Max) Meta	37.1%	—	—	224K
28	Claude Sonnet 5.5 (High, Default Fallback) Anthropic	34.3%	$4.98	17m 25s	147K
29	Kimi K3.1 Pro Kimi	34.3%	$10.23	71m 15s	139K
30	GPT-6.1 Terra (High) OpenAI	33.3%	$0.62	9m 2s	38K
31	Muse Wonder (Max) Meta	25.7%	$22.18	36m 33s	267K
32	GPT-6.1 Terra (Medium) OpenAI	24.8%	$0.38	5m 57s	24.8K
33	Claude Opus 5.5 (Low, Default Fallback) Anthropic	24.3%	$2.81	7m 59s	51.9K
34	GPT-6.1 Terra (Non-reasoning) OpenAI	24.3%	$0.51	5m 8s	21.3K
35	Muse Wonder (High) Meta	21.9%	$14.79	27m 23s	199K
36	Claude Haiku 5.5 (Max) Anthropic	20.0%	$1.03	25m 1s	382K
37	Muse Ember (Max) Meta	20.0%	$0.43	8m 6s	221K
38	Claude Haiku 5.5 (Xhigh) Anthropic	18.1%	$0.65	18m 12s	258K
39	Muse Ember (Xhigh) Meta	17.6%	$0.40	7m 22s	203K
40	Claude Sonnet 5.5 (Medium, Default Fallback) Anthropic	16.2%	$2.14	8m 49s	73.7K
41	DeepSeek V4.1 Flash 1017 (Max) DeepSeek	12.9%	$1.98	38m 47s	321K
42	Tencent Hy4 Tencent	12.9%	$2.62	—	151K
43	Qwen3.8 Max (0902) Alibaba	11.9%	$26.53	119m 25s	366K
44	Grok 4.7 (Xhigh) SpaceXAI	11.4%	$19.50	47m 52s	184K
45	GPT-6.1 Sol (Low) OpenAI	11.0%	$0.24	2m 47s	9.42K
46	Claude Sonnet 5.5 (Low, Default Fallback) Anthropic	10.5%	$1.23	5m 39s	47.2K
47	DeepSeek V4.1 Flash 1017 (High) DeepSeek	10.5%	$1.60	31m 27s	267K
48	Gemini 3.8 Flash (Medium) Google	10.5%	$14.46	23m 34s	158K
49	Claude Haiku 5.5 (High) Anthropic	10.0%	$0.43	13m 1s	178K
50	Gemini 3.8 Flash (High) Google	10.0%	$16.06	26m 5s	180K

1
Claude Fable 5.2 (Max, Default Fallback)
Anthropic
73.3%
$45.21
73m 49s
455K
2

Données de Artificial Analysis

Command Palette

Terminal-Bench-Science 0.1

Distribution des scores

Classement