CritPt v2

CritPt v2 est la version révisée de CritPt, un benchmark conçu pour évaluer les LLM sur des tâches de raisonnement physique de niveau recherche, comprenant 70 défis de recherche composites. Les modèles résolvent les problèmes à l'aide d'un outil d'exécution de code dans un bac à sable Linux hors ligne.

Poids de l'Indice d'intelligence: 10%
Modèles: 110 actuel, 128 au total
Médiane: 33.5%
Plage: 0.0%–76.5%

Distribution des scores

Le plus bas: 0.0%
Médiane: 33.5%
Le plus élevé: 76.5%

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	76.5%	$6.35	20m 0s	123K
2	Claude Fable 5.2 (Xhigh, Default Fallback) Anthropic	73.5%	$2.47	7m 48s	48.3K
3	Claude Fable 5.2 (High, Default Fallback) Anthropic	73.0%	$1.05	3m 42s	20.7K
4	GPT-6 Astra (Medium) OpenAI	72.0%	$0.41	2m 27s	6.27K
5	GPT-6 Astra (High) OpenAI	71.5%	$0.71	4m 46s	11.2K
6	GPT-6 Astra (Xhigh) OpenAI	71.0%	$0.85	5m 48s	13.5K
7	GPT-6 Astra (Max) OpenAI	70.5%	$1.14	8m 5s	18.3K
8	GPT-6.1 Sol (Max) OpenAI	69.5%	$0.23	5m 35s	18.8K
9	GPT-6.1 Sol (Xhigh) OpenAI	69.5%	$0.17	4m 2s	13.5K
10	Gemini 4 Carbon (High) Google	69.0%	$3.91	26m 33s	131K
11	Gemini 4 Carbon (Medium) Google	69.0%	$3.09	23m 21s	103K
12	GPT-6.1 Sol (High) OpenAI	68.5%	$0.13	3m 8s	10.7K
13	Claude Opus 5.5 (Max, Default Fallback) Anthropic	68.0%	$2.58	13m 51s	97.2K
14	GPT-6.1 Terra (Max) OpenAI	68.0%	$0.25	8m 43s	38.4K
15	Claude Opus 5.5 (Xhigh, Default Fallback) Anthropic	67.5%	$0.97	5m 38s	36.6K
16	GPT-6.1 Sol (Medium) OpenAI	67.0%	$0.071	1m 34s	5.55K
17	Gemini 4 Carbon (Low) Google	65.0%	$3.06	26m 4s	101K
18	GPT-6 Astra (Low) OpenAI	63.0%	$0.24	1m 29s	3.52K
19	GPT-6.1 Sol (Low) OpenAI	62.5%	$0.043	57.8s	3.26K
20	Claude Fable 5.2 (Medium, Default Fallback) Anthropic	61.5%	$0.37	1m 17s	7.01K
21	GPT-6.1 Terra (High) OpenAI	61.5%	$0.069	2m 31s	10.6K
22	GPT-6.1 Terra (Xhigh) OpenAI	60.5%	$0.12	4m 10s	18K
23	Claude Fable 5.2 (Low, Default Fallback) Anthropic	60.0%	$0.31	1m 4s	5.96K
24	Claude Opus 5.5 (High, Default Fallback) Anthropic	59.5%	$0.38	2m 17s	14.4K
25	Claude Sonnet 5.5 (Max, Default Fallback) Anthropic	59.0%	$1.93	15m 18s	141K
26	Claude Sonnet 5.5 (Xhigh, Default Fallback) Anthropic	57.5%	$0.45	4m 9s	34.7K
27	GPT-6.1 Terra (Medium) OpenAI	56.5%	$0.04	1m 28s	6.08K
28	Muse Wonder Full (Max) Meta	56.5%	—	—	65.4K
29	Claude Opus 5.5 (Medium, Default Fallback) Anthropic	54.0%	$0.25	1m 31s	9.39K
30	Gemini 4 Argon (High) Google	52.5%	$2.35	17m 35s	71.5K
31	Claude Sonnet 5.5 (High, Default Fallback) Anthropic	51.5%	$0.18	1m 37s	13.7K
32	Muse Wonder (High) Meta	51.0%	$1.17	5m 57s	43.3K
33	GPT-6.1 Terra (Low) OpenAI	50.5%	$0.018	36.4s	2.47K
34	Muse Wonder (Max) Meta	50.5%	$1.81	9m 5s	66.4K
35	Kimi K3.1 Pro Kimi	49.5%	$0.70	24m 34s	47.8K
36	Claude Haiku 5.5 (Xhigh) Anthropic	47.5%	$0.059	6m 4s	86.1K
37	Gemini 3.8 Flash (High) Google	46.5%	$1.15	9m 39s	66.5K
38	Claude Haiku 5.5 (High) Anthropic	45.0%	$0.033	3m 31s	48.1K
39	Gemini 3.8 Flash (Medium) Google	45.0%	$0.83	7m 27s	50.1K
40	Claude Haiku 5.5 (Max) Anthropic	43.5%	$0.10	9m 42s	148K
41	Grok 4.7 (Low) SpaceXAI	43.5%	$0.79	11m 40s	45.2K
42	GPT-6 Luna (Max) OpenAI	42.5%	$0.032	7m 26s	48.8K
43	Muse Ember (Max) Meta	39.0%	$0.054	2m 57s	80.7K
44	Tencent Hy4 Tencent	39.0%	$0.31	—	78.6K
45	Ling 3.1 Flash 1019 InclusionAI	38.5%	$0.32	—	157K
46	Muse Ember (Xhigh) Meta	38.5%	$0.041	2m 14s	61.3K
47	GPT-6 Luna (Xhigh) OpenAI	37.0%	$0.013	3m 11s	19.6K
48	Grok 4.7 (High) SpaceXAI	37.0%	$1.89	22m 38s	91.4K
49	Claude Opus 5.5 (Low, Default Fallback) Anthropic	36.0%	$0.12	40.5s	4.39K
50	Inkling 2 Small Thinking Machines	35.0%	$0.25	17m 44s	136K

1
Claude Fable 5.2 (Max, Default Fallback)
Anthropic
76.5%
$6.35
20m 0s
123K
2

Données de Artificial Analysis

Command Palette

CritPt v2

Distribution des scores

Classement