Ranking

Ranking: Contexto longo

Busca e raciocínio em entradas muito longas: MRCR v2 e tamanho de janela.

Geral738 Código437 Matemática56 Raciocínio451 Agentes55 Conhecimento450 Multimodal420 Agentes de código55 Raciocínio científico82 Multilíngue12 Escrita61 Chat e assistentes61 Pesquisa85 Chamada de ferramentas32 Contexto longo82 Velocidade440 Custo-benefício439 Pesos abertos173

Modelos: 82

#	Modelo	Podium Score	MRCR v2	Velocidade
01	GPT-6 Astra OpenAI · raciocínio	100.0	93%	82 t/s
02	GPT-5.6 Sol OpenAI	100.0	91.5%	72 t/s
03	GPT-5.4 OpenAI	100.0	—	60 t/s
4	Claude Fable 5 Anthropic · raciocínio	99.2	—	71 t/s
5	Kimi K3 Moonshot AI · Aberto · raciocínio	99.2	—	37 t/s
6	Claude Opus 5 Anthropic	99.2	—	60 t/s
7	Qwen3.8 Max Alibaba	99.2	—	55 t/s
8	Claude Opus 4.8 Anthropic	99.2	—	50 t/s
9	Muse Spark 1.1 Meta	99.2	—	208 t/s
10	Gemini 3.6 Flash Google	99.2	—	233 t/s
11	Claude Opus 4.7 Anthropic	99.2	—	80 t/s
12	Qwen3.7 Max Alibaba	99.2	—	203 t/s
13	Gemini 3.5 Flash Google	99.2	—	267 t/s
14	MiMo V2.5 Pro Xiaomi · Aberto	99.2	—	65 t/s
15	DeepSeek V4 Pro DeepSeek · Aberto	99.2	—	66 t/s
16	Claude Sonnet 5 Anthropic	99.2	—	96 t/s
17	Qwen3.7 Plus Alibaba	99.2	—	54 t/s
18	DeepSeek V4 Flash DeepSeek · Aberto	99.2	—	113 t/s
19	Gemini 3.5 Flash-Lite Google	99.2	—	389 t/s
20	Claude Sonnet 4.6 Anthropic	99.2	—	50 t/s
21	Qwen3.6 Plus Alibaba	99.2	—	55 t/s
22	GPT-5.6 Terra OpenAI	98.4	89.6%	127 t/s
23	DeepSeek V4.1 Flash DeepSeek · Aberto · raciocínio	98.0	—	214 t/s
24	Agnes 3.0 Flash Agnes · raciocínio	95.0	—	235 t/s
25	Inkling Small Thinking Machines	93.6	—	130 t/s
26	MiniMax M3 MiniMax	93.6	—	93 t/s
27	Inkling Thinking Machines	93.6	—	81 t/s
28	Nemotron 3 Ultra 550B NVIDIA · Aberto	93.4	—	142 t/s
29	Grok 4.5 xAI	93.2	—	59 t/s
30	Grok 4.3 xAI	93.2	—	129 t/s
31	GPT-5.3 Codex OpenAI	91.2	—	118 t/s
32	GPT-5.1 OpenAI	91.2	—	60 t/s
33	GPT-5 OpenAI	91.2	—	60 t/s
34	DeepSeek V4 Flash Vision DeepSeek	90.0	—	120 t/s
35	GLM-5.3 Zhipu AI	90.0	—	90 t/s
36	LongCat 2.0 LongCat · Aberto	90.0	—	42 t/s
37	GLM-5.3 Flash Z.ai · Aberto	88.0	—	182 t/s
38	GLM-5.2 Z.ai · Aberto	87.5	—	164 t/s
39	Kimi K2.6 Moonshot AI · Aberto	87.5	—	40 t/s
40	Nex N2 Pro Nex AGI	87.5	—	138 t/s
41	Kimi K2.7 Code Moonshot AI · Aberto	87.5	—	39 t/s
42	Hunyuan Hy3 Tencent	87.5	—	69 t/s
43	Gemma 4 31B Google · Aberto	87.5	—	35 t/s
44	Qwen3.5 397B-A17B Alibaba · Aberto	87.5	—	71 t/s
45	Qwen3.5 122B-A10B Alibaba · Aberto	87.5	—	126 t/s
46	Qwen3 Max Alibaba · Aberto	87.5	—	85 t/s
47	Claude Opus 4.6 Anthropic	86.3	76%	50 t/s
48	Qwen3.8 Flash Next Alibaba · Aberto	86.0	—	168 t/s
49	Claude Opus 4.5 Anthropic	85.2	—	50 t/s
50	Claude Sonnet 4.5 Anthropic	85.2	—	80 t/s
51	o3 OpenAI · raciocínio	85.2	—	90 t/s
52	Claude Opus 4.1 Anthropic	85.2	—	80 t/s
53	GPT-5.5 OpenAI	84.9	74%	60 t/s
54	DeepSeek V3.2 DeepSeek · Aberto	83.5	—	60 t/s
55	DeepSeek R1 DeepSeek · Aberto · raciocínio	83.5	—	60 t/s
56	GPT-5.2 OpenAI	81.3	—	90 t/s
57	Grok 4.6 (low) xAI	75.0	—	59 t/s
58	Grok 4.6 (medium) xAI	75.0	—	59 t/s
59	Grok 4.6 (xhigh) xAI	75.0	—	62 t/s
60	KAT-Coder-Pro V2 Kwai	75.0	—	103 t/s
61	Qwen3.8 27B (xhigh) Alibaba · Aberto	75.0	—	53 t/s
62	Qwen3.8 27B (low) Alibaba · Aberto	75.0	—	64 t/s
63	Qwen3.8 27B (medium) Alibaba · Aberto	75.0	—	62 t/s
64	Qwen3.8 27B (Non-reasoning) Alibaba · Aberto · raciocínio	75.0	—	62 t/s
65	Granite 4.2 30B Instruct IBM · Aberto	70.0	—	78 t/s
66	Granite 4.2 8B Instruct IBM · Aberto	68.0	—	115 t/s
67	Granite 4.2 3B Instruct IBM · Aberto	65.0	—	148 t/s
68	Agnes 2.5 Pro Beta Agnes	62.0	—	65 t/s
69	GPT-5.6 Luna OpenAI	56.6	41.3%	175 t/s
70	G9v3-39A5B AI9Stars · Aberto	50.0	—	60 t/s
71	G9v3-3B AI9Stars · Aberto	50.0	—	60 t/s
72	HyperNova 60B Multiverse · Aberto	50.0	—	367 t/s
73	LFM2.5-2.6B Liquid AI · Aberto	50.0	—	60 t/s
74	LFM2.5-8B-A1B Liquid AI · Aberto	50.0	—	338 t/s
75	LFM2.5-VL-1.6B Liquid AI · Aberto	50.0	—	395 t/s
76	Magistral Small 1.2 Mistral · Aberto	50.0	—	60 t/s
77	Gemini 3.1 Pro Google	43.3	26.3%	136 t/s
78	Gemini 3 Pro Google	43.3	26.3%	120 t/s
79	Gemini 3 Flash Google	39.7	22.1%	120 t/s
80	Seed 2.0 Pro ByteDance	13.8	—	40 t/s
81	ERNIE 5.1 Baidu	11.0	—	60 t/s
82	Muse Spark Meta	0.0	—	80 t/s

O Podium Score é uma mistura ponderada de cinco leaderboards. Saiba mais na metodologia.