Carregando...

Ranking de benchmarks de OCR

97 modelos de OCR medidos em 20 páginas escaneadas que quebram o OCR comum: digitalizações degradadas, matemática escaneada, tabelas e layouts em várias colunas. A medição do corpus difícil inclui 141 testes de regras; a Qualidade também inclui a página em branco quando há dados.

97 modelos classificados

20 páginas difíceis · 141 testes de regras

Publicado em Oct 5, 2026

Como medimos

Classificação geral

Uma linha por modelo, ordenada pela pontuação de qualidade. Mostramos a oferta de maior pontuação e o provedor onde o resultado foi medido.

97 modelos classificados · clique em uma coluna para ordenar

Posição	Modelo
1	Mistral OCR 4.1 via Mistral	84.0±6.3	69	80	92	95	—	200 pages / $1$0.005/page	100
2	Datalab Convert (accurate) via Datalab	77.2±7.1	54	73	92	91	25.9s	100 pages / $1$0.01/page	100
3	Ministral 8B 2512 via Mistral	69.7±7.7	62	67	84	67	2.3s	~7,040 pages / $1$0.000142/página · Referência de benchmark	100
4	Ministral 3 8B 2512 Mistral · via OpenRouter	68.3±7.8	62	61	84	67	1.9s	~6,670 pages / $1$0.00015/página · Referência de benchmark	100
5	Claude Opus 4.5 Anthropic · via OpenRouter	67.5±7.5	58	78	72	62	5.6s	~46.1 pages / $1~$0.0217 per standard page	100
6	Claude Opus 4.8 (Fast) Anthropic · via OpenRouter	66.9±8.2	65	33	88	81	2.5s	~27.3 pages / $1~$0.0367 per standard page	—
7	Qwen3.8 Flash Qwen · via OpenRouter	65.5±7.7	69	77	40	76	—	~2,200 pages / $1~$0.000454 per standard page	100
8	Muse Spark 1.2 Meta · via OpenRouter	65.5±7.8	73	70	48	71	7.6s	~103 pages / $1~$0.00975 per standard page	—
9	Mistral Medium 3 Mistral · via OpenRouter	64.9±8.0	62	59	72	67	3.9s	~684 pages / $1~$0.00146 per standard page	100
10	GPT-5.6 Luna OpenAI · via OpenRouter	64.8±8.1	73	52	48	86	4.8s	~816 pages / $1~$0.00123 per standard page	100
11	Mistral Medium 3.1 Mistral · via OpenRouter	64.6±8.1	73	51	68	67	2.2s	~684 pages / $1~$0.00146 per standard page	100
12	Step 3.7 Flash StepFun · via OpenRouter	63.6±8.0	69	59	64	62	11.9s	~1,150 pages / $1~$0.000867 per standard page	100
13	Gemini 3 Flash Preview Google · via OpenRouter	63.6±7.7	65	78	44	67	2.7s	~750 pages / $1$0.001333/página · Referência de benchmark	100
14	Claude Opus 4.6 Anthropic · via OpenRouter	63.4±8.1	65	46	80	62	6.1s	~46.1 pages / $1~$0.0217 per standard page	100
15	Claude Opus 4 Anthropic · via OpenRouter	63.1±8.2	58	46	72	76	8.0s	~18.3 pages / $1~$0.0547 per standard page	—
16	Gemini Flash Latest via Google	63.0±7.7	69	78	52	52	6.1s	~1,660 pages / $1$0.000601/página · Referência de benchmark	100
17	Qwen3 VL 235B A22B Instruct Qwen · via OpenRouter	63.0±8.2	69	45	52	86	4.4s	~678 pages / $1~$0.00148 per standard page	100
18	Mistral Medium 2505 via Mistral	62.1±8.0	62	61	64	62	3.7s	~739 pages / $1~$0.00135 per standard page	100
19	Mistral Medium 2508 via Mistral	61.9±8.2	77	46	72	52	2.0s	~602 pages / $1~$0.00166 per standard page	100
20	Claude Sonnet 4.5 via Anthropic	61.1±8.3	69	30	64	81	5.7s	~211 pages / $1$0.004746/página · Referência de benchmark	100
21	Llama 4 Maverick Meta · via OpenRouter	61.0±7.4	69	74	60	81	4.9s	~1,560 pages / $1~$0.00064 per standard page	33
22	Mistral Large 3 2512 Mistral · via OpenRouter	58.5±8.3	69	35	68	62	4.6s	~849 pages / $1~$0.00118 per standard page	100
23	Ministral 14B 2512 via Mistral	57.7±8.2	65	26	68	71	4.9s	~7,520 pages / $1$0.000133/página · Referência de benchmark	100
24	MiMo-V2.5 Xiaomi · via OpenRouter	57.3±8.2	58	54	56	62	13.3s	~2,620 pages / $1~$0.000381 per standard page	100
25	Gemini 3.5 Flash-Lite Google · via OpenRouter	56.9±8.2	69	57	16	86	1.7s	~504 pages / $1~$0.00198 per standard page	100
26	Magistral Medium 2509 via Mistral	56.6±8.2	65	16	64	81	1.9s	~334 pages / $1$0.002996/página · Referência de benchmark	100
27	Grok 4.5 xAI · via OpenRouter	56.5±8.2	73	54	28	71	5.3s	~176 pages / $1~$0.00567 per standard page	100
28	Gemini 3.6 Flash via Google	56.5±8.3	69	42	48	67	4.1s	~179 pages / $1~$0.00559 per standard page	100
29	Ministral 3B 2512 via Mistral	56.2±8.2	58	44	76	48	1.7s	~1,170 pages / $1$0.000856/página · Referência de benchmark	100
30	Qwen3-VL 30B A3B Instruct (DeepInfra) via DeepInfra	56.1±8.2	73	62	8	81	4.5s	~2,470 pages / $1~$0.000404 per standard page	100
31	Mistral Medium 3.5.0 via Mistral	55.4±8.2	65	16	64	76	—	—Referência de benchmark	—
32	MiniMax M3 MiniMax · via OpenRouter	55.0±8.3	69	38	56	57	4.8s	~940 pages / $1~$0.00106 per standard page	100
33	GLM 5.3 Flash Z.ai · via OpenRouter	55.0±8.0	62	0	68	91	—	~1,420 pages / $1~$0.000703 per standard page	100
34	Ministral 3 3B 2512 Mistral · via OpenRouter	54.5±8.3	58	41	72	48	1.8s	~10,300 pages / $1$0.000097/página · Referência de benchmark	100
35	Grok 4.3 xAI · via OpenRouter	54.3±8.1	69	12	60	76	4.7s	~318 pages / $1~$0.00314 per standard page	100
36	Claude Sonnet 4.6 Anthropic · via OpenRouter	54.1±8.2	73	17	64	62	5.2s	~76.9 pages / $1~$0.013 per standard page	100
37	GPT-5.6 Luna Pro OpenAI · via OpenRouter	53.9±8.2	69	57	28	62	22.0s	~189 pages / $1~$0.0053 per standard page	100
38	Llama 4 Scout Meta · via OpenRouter	53.6±8.2	54	67	32	62	1.6s	~1,370 pages / $1~$0.000731 per standard page	100
39	Mistral Small 2506 via Mistral	53.6±8.1	65	16	76	57	1.9s	~7,250 pages / $1$0.000138/página · Referência de benchmark	100
40	Gemini Flash-Lite Latest via Google	53.4±8.1	73	70	4	67	1.6s	~4,930 pages / $1$0.000203/página · Referência de benchmark	100
41	Seed-2.0-Code ByteDance Seed · via OpenRouter	53.0±8.1	73	75	16	48	15.4s	~259 pages / $1~$0.00385 per standard page	100
42	Magistral Small 2509 via Mistral	52.9±8.1	65	13	76	57	2.0s	~1,340 pages / $1$0.000749/página · Referência de benchmark	100
43	Gemini 3.7 Flash Google · via OpenRouter	51.6±8.3	69	45	40	52	3.8s	~226 pages / $1~$0.00442 per standard page	100
44	Qwen2.5-VL 32B Instruct (DeepInfra) via DeepInfra	51.5±8.2	62	58	20	67	5.2s	~467 pages / $1$0.00214/page	100
45	Claude Fable 5 Anthropic · via OpenRouter	50.8±7.9	69	0	72	62	6.8s	~27.3 pages / $1~$0.0367 per standard page	—
46	Claude Sonnet 4 Anthropic · via OpenRouter	49.5±8.0	69	16	32	81	3.8s	~76.9 pages / $1~$0.013 per standard page	100
47	Muse Glimmer 30B Meta · via OpenRouter	49.1±8.1	77	28	16	76	7.6s	~754 pages / $1~$0.00133 per standard page	100
48	Claude Sonnet 5 Anthropic · via OpenRouter	47.4±7.7	69	0	44	76	4.6s	~136 pages / $1~$0.00733 per standard page	100
49	Qwen3.8 27B Qwen · via OpenRouter	47.3±8.0	69	17	36	67	5.2s	~455 pages / $1~$0.0022 per standard page	100
50	Claude Opus 5 (Fast) via OpenRouter	46.0±7.7	65	0	52	67	3.8s	~27.3 pages / $1~$0.0367 per standard page	—
51	Gemini 3.1 Flash Lite via Google	45.6±8.3	65	28	64	86	1.2s	~821 pages / $1~$0.00122 per standard page	0
52	Claude Opus 5 via OpenRouter	45.5±7.7	65	3	52	62	5.0s	~45.2 pages / $1~$0.0221 per standard page	100
53	Gemini 3.1 Flash Lite Preview via Google	45.5±8.2	69	42	64	67	1.1s	~845 pages / $1~$0.00118 per standard page	0
54	Grok Build 0.1 xAI · via OpenRouter	45.1±7.6	77	0	32	71	32.7s	~89.5 pages / $1~$0.0112 per standard page	100
55	Gemini 3.5 Flash Google · via OpenRouter	44.2±8.0	73	22	44	38	5.9s	~59.6 pages / $1~$0.0168 per standard page	—
56	Qwen3.6 35B A3B Qwen · via OpenRouter	43.8±8.1	69	39	0	67	6.4s	~1,170 pages / $1~$0.000852 per standard page	100
57	GPT-4.1 Nano OpenAI · via OpenRouter	43.5±7.9	58	22	28	67	2.3s	~3,040 pages / $1~$0.000329 per standard page	100
58	Nex-N2-Mini Nex AGI · via OpenRouter	43.1±8.0	73	25	8	67	2.3s	~8,640 pages / $1~$0.000116 per standard page	—
59	Gemini 3.1 Flash Image Preview (Gemini 3.1 Flash Image Preview) Google · via OpenRouter	42.8±8.3	73	38	48	52	1.8s	~1,140 pages / $1$0.000878/página · Referência de benchmark	33
60	Claude Haiku 4.5 Anthropic · via OpenRouter	42.6±7.9	69	22	32	48	3.4s	~274 pages / $1~$0.00365 per standard page	100
61	o4 Mini via OpenAI	41.7±7.4	73	0	8	86	11.7s	~72.1 pages / $1~$0.0139 per standard page	—
62	Nova 2 Lite Amazon · via OpenRouter	41.6±7.8	69	16	48	33	2.0s	~580 pages / $1~$0.00172 per standard page	100
63	Perceptron Mk1 Perceptron · via OpenRouter	41.1±7.4	62	1	16	86	13.6s	~982 pages / $1~$0.00102 per standard page	100
64	Mistral Large 2512 via Mistral	40.5±8.2	73	25	72	52	4.3s	~376 pages / $1$0.00266/página · Referência de benchmark	0
65	Nova Lite 1.0 Amazon · via OpenRouter	40.1±7.4	65	0	52	43	2.2s	~4,260 pages / $1~$0.000235 per standard page	100
66	Qwen3.6 27B Qwen · via OpenRouter	40.0±8.1	69	35	0	76	10.4s	~365 pages / $1~$0.00274 per standard page	67
67	Qwen3.6 Flash Qwen · via OpenRouter	39.0±8.2	73	42	0	81	15.4s	~1,270 pages / $1~$0.000789 per standard page	33
68	Anthropic Claude Haiku Latest via OpenRouter	37.9±7.7	69	19	16	48	—	~231 pages / $1~$0.00433 per standard page	100
69	GPT-5.6 Terra Pro OpenAI · via OpenRouter	37.6±7.5	69	15	0	67	14.2s	~50.7 pages / $1~$0.0197 per standard page	—
70	Sonar Pro Perplexity · via OpenRouter	37.1±7.6	69	17	0	62	3.1s	~91.9 pages / $1~$0.0109 per standard page	100
71	GPT 4o (2024-11-20) via OpenAI	36.3±7.5	77	16	0	52	1.8s	~263 pages / $1$0.003798/página · Referência de benchmark	100
72	GPT-5.6 Sol OpenAI · via OpenRouter	36.1±7.1	73	0	0	71	6.4s	~139 pages / $1~$0.00721 per standard page	100
73	GPT 4.1 (2025-04-14) via OpenAI	36.1±7.3	65	6	16	57	1.6s	~333 pages / $1$0.003006/página · Referência de benchmark	100
74	Grok 4.6 SpaceXAI · via OpenRouter	36.0±7.8	69	19	0	76	4.0s	~99.4 pages / $1~$0.0101 per standard page	67
75	Azure Document Intelligence v4 via Azure	35.1±7.0	69	0	0	71	1.4s	~93.5 pages / $1$0.0107/página · Referência de benchmark	100
76	GPT-5.6 Sol Pro OpenAI · via OpenRouter	33.7±7.6	58	25	0	52	8.6s	~77.3 pages / $1~$0.0129 per standard page	—
77	GPT-5.1-Codex OpenAI · via OpenRouter	33.5±6.9	77	0	0	57	6.0s	~135 pages / $1~$0.0074 per standard page	100
78	Sonar Perplexity · via OpenRouter	32.8±6.8	69	0	0	62	2.5s	~966 pages / $1~$0.00104 per standard page	100
79	Sonar Pro Search Perplexity · via OpenRouter	32.7±7.3	65	13	0	52	4.2s	~203 pages / $1$0.004935/página · Referência de benchmark	—
80	GPT 4o Mini (2024-07-18) via OpenAI	32.3±6.8	77	0	0	52	2.2s	~254 pages / $1$0.003933/página · Referência de benchmark	100
81	Document AI via Google	32.0±6.8	62	0	0	67	622ms	~100 pages / $1$0.01/página · Referência de benchmark	100
82	OpenAI GPT Mini Latest via OpenRouter	31.8±6.8	65	0	0	62	—	~313 pages / $1~$0.00319 per standard page	100
83	GPT-5.4 Image 2 OpenAI · via OpenRouter	31.6±6.8	69	0	0	57	2.5s	~48.2 pages / $1~$0.0207 per standard page	100
84	GPT 4.1 Mini (2025-04-14) via OpenAI	31.1±7.7	69	0	44	71	2.4s	~1,660 pages / $1$0.000602/página · Referência de benchmark	0
85	Gemini 3.1 Pro Preview Custom Tools via Google	30.4±6.7	69	0	0	52	27.9s	~18.3 pages / $1~$0.0546 per standard page	—
86	o3 via OpenAI	30.4±6.7	69	0	0	52	5.8s	~54.7 pages / $1~$0.0183 per standard page	100
87	Gemini 2.5 Flash Google · via OpenRouter	30.1±7.0	69	0	0	71	2.1s	~974 pages / $1$0.001027/página · Referência de benchmark	67
88	GPT-5.6 Terra OpenAI · via OpenRouter	29.4±6.6	65	0	0	52	2.6s	~118 pages / $1~$0.00851 per standard page	100
89	GPT-5.5 OpenAI · via OpenRouter	29.2±6.6	69	0	0	48	7.6s	~35.6 pages / $1~$0.0281 per standard page	—
90	Inkling Small Thinking Machines · via OpenRouter	28.4±7.5	73	0	48	52	2.5s	~389 pages / $1~$0.00257 per standard page	0
91	Claude 3 Haiku Anthropic · via OpenRouter	27.7±6.4	54	0	0	57	2.7s	~1,100 pages / $1~$0.000912 per standard page	—
92	Gemini 2.5 Pro via Google	25.6±6.3	69	0	0	33	27.0s	~25.1 pages / $1~$0.0398 per standard page	—
93	Nova Pro 1.0 Amazon · via OpenRouter	21.1±5.9	65	0	0	19	2.4s	~320 pages / $1~$0.00313 per standard page	100
94	AWS Textract via AWS	19.5±5.8	73	0	0	5	1.2s	~623 pages / $1$0.001605/page	100
95	docTR via OpenOCR (Local)	12.8±4.8	46	0	0	5	1.1s	500 pages / $1$0.002/page	100
96	UI-TARS 7B ByteDance · via OpenRouter	11.5±6.4	58	10	4	14	3.5s	~4,180 pages / $1~$0.000239 per standard page	33
97	EasyOCR via OpenOCR (Local)	4.8±3.0	19	0	0	0	1.8s	500 pages / $1$0.002/page	100

Metodologia

As posições seguem a pontuação de qualidade sem arredondamento, da maior para a menor. Pontuações idênticas usam a ordem do ID canônico para atribuir posições únicas. O corpus contém 20 páginas digitalizadas do olmOCR-bench da AI2, avaliadas com 141 testes de regras. A Qualidade mostra uma porcentagem; seu intervalo de confiança de 95% vem do corpus difícil, portanto posições próximas podem ter incertezas sobrepostas. As categorias mostram taxas de aprovação. Velocidade é o tempo médio por página via OpenOCR, incluindo a rede. O preço mostra páginas por $1 e custo por página; preços por tokens são estimativas. As recomendações de velocidade e preço vêm dos 10 primeiros modelos. Quando várias ofertas atendem ao mesmo modelo, mostramos a de maior pontuação. Quando publicados, os resultados de um corpus separado de 11 imagens de documentos limpos aparecem na página de evidências. Eles não determinam a classificação. Modelos medidos apenas no corpus difícil não têm pontuação nem latência do corpus limpo. Página em branco: cada modelo também recebe uma página em branco três vezes. Um modelo que inventa texto perde até 15 pontos de qualidade, proporcionalmente à frequência com que o faz. Leia a metodologia completa →

Ranking de benchmarks de OCR

97 modelos classificados

20 páginas difíceis · 141 testes de regras

Publicado em Oct 5, 2026

Como medimos

Classificação geral

Uma linha por modelo, ordenada pela pontuação de qualidade. Mostramos a oferta de maior pontuação e o provedor onde o resultado foi medido.

97 modelos classificados · clique em uma coluna para ordenar

Posição	Modelo
1	Mistral OCR 4.1 via Mistral	84.0±6.3	69	80	92	95	—	200 pages / $1$0.005/page	100
2	Datalab Convert (accurate) via Datalab	77.2±7.1	54	73	92	91	25.9s	100 pages / $1$0.01/page	100
3	Ministral 8B 2512 via Mistral	69.7±7.7	62	67	84	67	2.3s	~7,040 pages / $1$0.000142/página · Referência de benchmark	100
4	Ministral 3 8B 2512 Mistral · via OpenRouter	68.3±7.8	62	61	84	67	1.9s	~6,670 pages / $1$0.00015/página · Referência de benchmark	100
5	Claude Opus 4.5 Anthropic · via OpenRouter	67.5±7.5	58	78	72	62	5.6s	~46.1 pages / $1~$0.0217 per standard page	100
6	Claude Opus 4.8 (Fast) Anthropic · via OpenRouter	66.9±8.2	65	33	88	81	2.5s	~27.3 pages / $1~$0.0367 per standard page	—
7	Qwen3.8 Flash Qwen · via OpenRouter	65.5±7.7	69	77	40	76	—	~2,200 pages / $1~$0.000454 per standard page	100
8	Muse Spark 1.2 Meta · via OpenRouter	65.5±7.8	73	70	48	71	7.6s	~103 pages / $1~$0.00975 per standard page	—
9	Mistral Medium 3 Mistral · via OpenRouter	64.9±8.0	62	59	72	67	3.9s	~684 pages / $1~$0.00146 per standard page	100
10	GPT-5.6 Luna OpenAI · via OpenRouter	64.8±8.1	73	52	48	86	4.8s	~816 pages / $1~$0.00123 per standard page	100
11	Mistral Medium 3.1 Mistral · via OpenRouter	64.6±8.1	73	51	68	67	2.2s	~684 pages / $1~$0.00146 per standard page	100
12	Step 3.7 Flash StepFun · via OpenRouter	63.6±8.0	69	59	64	62	11.9s	~1,150 pages / $1~$0.000867 per standard page	100
13	Gemini 3 Flash Preview Google · via OpenRouter	63.6±7.7	65	78	44	67	2.7s	~750 pages / $1$0.001333/página · Referência de benchmark	100
14	Claude Opus 4.6 Anthropic · via OpenRouter	63.4±8.1	65	46	80	62	6.1s	~46.1 pages / $1~$0.0217 per standard page	100
15	Claude Opus 4 Anthropic · via OpenRouter	63.1±8.2	58	46	72	76	8.0s	~18.3 pages / $1~$0.0547 per standard page	—
16	Gemini Flash Latest via Google	63.0±7.7	69	78	52	52	6.1s	~1,660 pages / $1$0.000601/página · Referência de benchmark	100
17	Qwen3 VL 235B A22B Instruct Qwen · via OpenRouter	63.0±8.2	69	45	52	86	4.4s	~678 pages / $1~$0.00148 per standard page	100
18	Mistral Medium 2505 via Mistral	62.1±8.0	62	61	64	62	3.7s	~739 pages / $1~$0.00135 per standard page	100
19	Mistral Medium 2508 via Mistral	61.9±8.2	77	46	72	52	2.0s	~602 pages / $1~$0.00166 per standard page	100
20	Claude Sonnet 4.5 via Anthropic	61.1±8.3	69	30	64	81	5.7s	~211 pages / $1$0.004746/página · Referência de benchmark	100
21	Llama 4 Maverick Meta · via OpenRouter	61.0±7.4	69	74	60	81	4.9s	~1,560 pages / $1~$0.00064 per standard page	33
22	Mistral Large 3 2512 Mistral · via OpenRouter	58.5±8.3	69	35	68	62	4.6s	~849 pages / $1~$0.00118 per standard page	100
23	Ministral 14B 2512 via Mistral	57.7±8.2	65	26	68	71	4.9s	~7,520 pages / $1$0.000133/página · Referência de benchmark	100
24	MiMo-V2.5 Xiaomi · via OpenRouter	57.3±8.2	58	54	56	62	13.3s	~2,620 pages / $1~$0.000381 per standard page	100
25	Gemini 3.5 Flash-Lite Google · via OpenRouter	56.9±8.2	69	57	16	86	1.7s	~504 pages / $1~$0.00198 per standard page	100
26	Magistral Medium 2509 via Mistral	56.6±8.2	65	16	64	81	1.9s	~334 pages / $1$0.002996/página · Referência de benchmark	100
27	Grok 4.5 xAI · via OpenRouter	56.5±8.2	73	54	28	71	5.3s	~176 pages / $1~$0.00567 per standard page	100
28	Gemini 3.6 Flash via Google	56.5±8.3	69	42	48	67	4.1s	~179 pages / $1~$0.00559 per standard page	100
29	Ministral 3B 2512 via Mistral	56.2±8.2	58	44	76	48	1.7s	~1,170 pages / $1$0.000856/página · Referência de benchmark	100
30	Qwen3-VL 30B A3B Instruct (DeepInfra) via DeepInfra	56.1±8.2	73	62	8	81	4.5s	~2,470 pages / $1~$0.000404 per standard page	100
31	Mistral Medium 3.5.0 via Mistral	55.4±8.2	65	16	64	76	—	—Referência de benchmark	—
32	MiniMax M3 MiniMax · via OpenRouter	55.0±8.3	69	38	56	57	4.8s	~940 pages / $1~$0.00106 per standard page	100
33	GLM 5.3 Flash Z.ai · via OpenRouter	55.0±8.0	62	0	68	91	—	~1,420 pages / $1~$0.000703 per standard page	100
34	Ministral 3 3B 2512 Mistral · via OpenRouter	54.5±8.3	58	41	72	48	1.8s	~10,300 pages / $1$0.000097/página · Referência de benchmark	100
35	Grok 4.3 xAI · via OpenRouter	54.3±8.1	69	12	60	76	4.7s	~318 pages / $1~$0.00314 per standard page	100
36	Claude Sonnet 4.6 Anthropic · via OpenRouter	54.1±8.2	73	17	64	62	5.2s	~76.9 pages / $1~$0.013 per standard page	100
37	GPT-5.6 Luna Pro OpenAI · via OpenRouter	53.9±8.2	69	57	28	62	22.0s	~189 pages / $1~$0.0053 per standard page	100
38	Llama 4 Scout Meta · via OpenRouter	53.6±8.2	54	67	32	62	1.6s	~1,370 pages / $1~$0.000731 per standard page	100
39	Mistral Small 2506 via Mistral	53.6±8.1	65	16	76	57	1.9s	~7,250 pages / $1$0.000138/página · Referência de benchmark	100
40	Gemini Flash-Lite Latest via Google	53.4±8.1	73	70	4	67	1.6s	~4,930 pages / $1$0.000203/página · Referência de benchmark	100
41	Seed-2.0-Code ByteDance Seed · via OpenRouter	53.0±8.1	73	75	16	48	15.4s	~259 pages / $1~$0.00385 per standard page	100
42	Magistral Small 2509 via Mistral	52.9±8.1	65	13	76	57	2.0s	~1,340 pages / $1$0.000749/página · Referência de benchmark	100
43	Gemini 3.7 Flash Google · via OpenRouter	51.6±8.3	69	45	40	52	3.8s	~226 pages / $1~$0.00442 per standard page	100
44	Qwen2.5-VL 32B Instruct (DeepInfra) via DeepInfra	51.5±8.2	62	58	20	67	5.2s	~467 pages / $1$0.00214/page	100
45	Claude Fable 5 Anthropic · via OpenRouter	50.8±7.9	69	0	72	62	6.8s	~27.3 pages / $1~$0.0367 per standard page	—
46	Claude Sonnet 4 Anthropic · via OpenRouter	49.5±8.0	69	16	32	81	3.8s	~76.9 pages / $1~$0.013 per standard page	100
47	Muse Glimmer 30B Meta · via OpenRouter	49.1±8.1	77	28	16	76	7.6s	~754 pages / $1~$0.00133 per standard page	100
48	Claude Sonnet 5 Anthropic · via OpenRouter	47.4±7.7	69	0	44	76	4.6s	~136 pages / $1~$0.00733 per standard page	100
49	Qwen3.8 27B Qwen · via OpenRouter	47.3±8.0	69	17	36	67	5.2s	~455 pages / $1~$0.0022 per standard page	100
50	Claude Opus 5 (Fast) via OpenRouter	46.0±7.7	65	0	52	67	3.8s	~27.3 pages / $1~$0.0367 per standard page	—
51	Gemini 3.1 Flash Lite via Google	45.6±8.3	65	28	64	86	1.2s	~821 pages / $1~$0.00122 per standard page	0
52	Claude Opus 5 via OpenRouter	45.5±7.7	65	3	52	62	5.0s	~45.2 pages / $1~$0.0221 per standard page	100
53	Gemini 3.1 Flash Lite Preview via Google	45.5±8.2	69	42	64	67	1.1s	~845 pages / $1~$0.00118 per standard page	0
54	Grok Build 0.1 xAI · via OpenRouter	45.1±7.6	77	0	32	71	32.7s	~89.5 pages / $1~$0.0112 per standard page	100
55	Gemini 3.5 Flash Google · via OpenRouter	44.2±8.0	73	22	44	38	5.9s	~59.6 pages / $1~$0.0168 per standard page	—
56	Qwen3.6 35B A3B Qwen · via OpenRouter	43.8±8.1	69	39	0	67	6.4s	~1,170 pages / $1~$0.000852 per standard page	100
57	GPT-4.1 Nano OpenAI · via OpenRouter	43.5±7.9	58	22	28	67	2.3s	~3,040 pages / $1~$0.000329 per standard page	100
58	Nex-N2-Mini Nex AGI · via OpenRouter	43.1±8.0	73	25	8	67	2.3s	~8,640 pages / $1~$0.000116 per standard page	—
59	Gemini 3.1 Flash Image Preview (Gemini 3.1 Flash Image Preview) Google · via OpenRouter	42.8±8.3	73	38	48	52	1.8s	~1,140 pages / $1$0.000878/página · Referência de benchmark	33
60	Claude Haiku 4.5 Anthropic · via OpenRouter	42.6±7.9	69	22	32	48	3.4s	~274 pages / $1~$0.00365 per standard page	100
61	o4 Mini via OpenAI	41.7±7.4	73	0	8	86	11.7s	~72.1 pages / $1~$0.0139 per standard page	—
62	Nova 2 Lite Amazon · via OpenRouter	41.6±7.8	69	16	48	33	2.0s	~580 pages / $1~$0.00172 per standard page	100
63	Perceptron Mk1 Perceptron · via OpenRouter	41.1±7.4	62	1	16	86	13.6s	~982 pages / $1~$0.00102 per standard page	100
64	Mistral Large 2512 via Mistral	40.5±8.2	73	25	72	52	4.3s	~376 pages / $1$0.00266/página · Referência de benchmark	0
65	Nova Lite 1.0 Amazon · via OpenRouter	40.1±7.4	65	0	52	43	2.2s	~4,260 pages / $1~$0.000235 per standard page	100
66	Qwen3.6 27B Qwen · via OpenRouter	40.0±8.1	69	35	0	76	10.4s	~365 pages / $1~$0.00274 per standard page	67
67	Qwen3.6 Flash Qwen · via OpenRouter	39.0±8.2	73	42	0	81	15.4s	~1,270 pages / $1~$0.000789 per standard page	33
68	Anthropic Claude Haiku Latest via OpenRouter	37.9±7.7	69	19	16	48	—	~231 pages / $1~$0.00433 per standard page	100
69	GPT-5.6 Terra Pro OpenAI · via OpenRouter	37.6±7.5	69	15	0	67	14.2s	~50.7 pages / $1~$0.0197 per standard page	—
70	Sonar Pro Perplexity · via OpenRouter	37.1±7.6	69	17	0	62	3.1s	~91.9 pages / $1~$0.0109 per standard page	100
71	GPT 4o (2024-11-20) via OpenAI	36.3±7.5	77	16	0	52	1.8s	~263 pages / $1$0.003798/página · Referência de benchmark	100
72	GPT-5.6 Sol OpenAI · via OpenRouter	36.1±7.1	73	0	0	71	6.4s	~139 pages / $1~$0.00721 per standard page	100
73	GPT 4.1 (2025-04-14) via OpenAI	36.1±7.3	65	6	16	57	1.6s	~333 pages / $1$0.003006/página · Referência de benchmark	100
74	Grok 4.6 SpaceXAI · via OpenRouter	36.0±7.8	69	19	0	76	4.0s	~99.4 pages / $1~$0.0101 per standard page	67
75	Azure Document Intelligence v4 via Azure	35.1±7.0	69	0	0	71	1.4s	~93.5 pages / $1$0.0107/página · Referência de benchmark	100
76	GPT-5.6 Sol Pro OpenAI · via OpenRouter	33.7±7.6	58	25	0	52	8.6s	~77.3 pages / $1~$0.0129 per standard page	—
77	GPT-5.1-Codex OpenAI · via OpenRouter	33.5±6.9	77	0	0	57	6.0s	~135 pages / $1~$0.0074 per standard page	100
78	Sonar Perplexity · via OpenRouter	32.8±6.8	69	0	0	62	2.5s	~966 pages / $1~$0.00104 per standard page	100
79	Sonar Pro Search Perplexity · via OpenRouter	32.7±7.3	65	13	0	52	4.2s	~203 pages / $1$0.004935/página · Referência de benchmark	—
80	GPT 4o Mini (2024-07-18) via OpenAI	32.3±6.8	77	0	0	52	2.2s	~254 pages / $1$0.003933/página · Referência de benchmark	100
81	Document AI via Google	32.0±6.8	62	0	0	67	622ms	~100 pages / $1$0.01/página · Referência de benchmark	100
82	OpenAI GPT Mini Latest via OpenRouter	31.8±6.8	65	0	0	62	—	~313 pages / $1~$0.00319 per standard page	100
83	GPT-5.4 Image 2 OpenAI · via OpenRouter	31.6±6.8	69	0	0	57	2.5s	~48.2 pages / $1~$0.0207 per standard page	100
84	GPT 4.1 Mini (2025-04-14) via OpenAI	31.1±7.7	69	0	44	71	2.4s	~1,660 pages / $1$0.000602/página · Referência de benchmark	0
85	Gemini 3.1 Pro Preview Custom Tools via Google	30.4±6.7	69	0	0	52	27.9s	~18.3 pages / $1~$0.0546 per standard page	—
86	o3 via OpenAI	30.4±6.7	69	0	0	52	5.8s	~54.7 pages / $1~$0.0183 per standard page	100
87	Gemini 2.5 Flash Google · via OpenRouter	30.1±7.0	69	0	0	71	2.1s	~974 pages / $1$0.001027/página · Referência de benchmark	67
88	GPT-5.6 Terra OpenAI · via OpenRouter	29.4±6.6	65	0	0	52	2.6s	~118 pages / $1~$0.00851 per standard page	100
89	GPT-5.5 OpenAI · via OpenRouter	29.2±6.6	69	0	0	48	7.6s	~35.6 pages / $1~$0.0281 per standard page	—
90	Inkling Small Thinking Machines · via OpenRouter	28.4±7.5	73	0	48	52	2.5s	~389 pages / $1~$0.00257 per standard page	0
91	Claude 3 Haiku Anthropic · via OpenRouter	27.7±6.4	54	0	0	57	2.7s	~1,100 pages / $1~$0.000912 per standard page	—
92	Gemini 2.5 Pro via Google	25.6±6.3	69	0	0	33	27.0s	~25.1 pages / $1~$0.0398 per standard page	—
93	Nova Pro 1.0 Amazon · via OpenRouter	21.1±5.9	65	0	0	19	2.4s	~320 pages / $1~$0.00313 per standard page	100
94	AWS Textract via AWS	19.5±5.8	73	0	0	5	1.2s	~623 pages / $1$0.001605/page	100
95	docTR via OpenOCR (Local)	12.8±4.8	46	0	0	5	1.1s	500 pages / $1$0.002/page	100
96	UI-TARS 7B ByteDance · via OpenRouter	11.5±6.4	58	10	4	14	3.5s	~4,180 pages / $1~$0.000239 per standard page	33
97	EasyOCR via OpenOCR (Local)	4.8±3.0	19	0	0	0	1.8s	500 pages / $1$0.002/page	100

Metodologia

Ranking de benchmarks de OCR

Classificação geral

Aguardando medição (23)

Aliases (7)

Metodologia

Ranking de benchmarks de OCR

Classificação geral

Aguardando medição (23)

Aliases (7)

Metodologia

Melhores da categoria

Classificação geral

Aguardando medição (23)

Aliases (7)

Metodologia

Melhores da categoria

Classificação geral

Aguardando medição (23)

Aliases (7)

Metodologia