Loading...
Loading...
Side-by-side comparison of available engines. Published benchmark scores and live operational metrics are shown when available.
Highest published score
Datalab Convert (accurate)
77.2%
Fastest
ScaleDown: Extract
190ms
Lowest reference cost
ScaleDown: Summarize
~28,900 pages / $1~$0.0000346 per standard page
Most Languages
Tesseract 5
116 languages
| Engine | Benchmark score ↓ | Latency (p50) | Price | Uptime | Languages | Strengths |
|---|---|---|---|---|---|---|
| Datalab Convert (accurate) | 77.2% | 6.7s | 100 pages / $1$0.01/page | 100% | — | |
| Meta: Llama 4 Maverick | 71% | 25.5s | ~1,560 pages / $1~$0.00064 per standard page | 94.44% | 100 | wide-model-selectionunified-apicost-effective |
| Anthropic: Claude Opus 4.5 | 67.5% | 2.6s | ~46.1 pages / $1~$0.0217 per standard page | 93.33% | 100 | wide-model-selectionunified-apicost-effective |
| Meta: Muse Spark 1.2 | 65.5% | 4.2s | ~103 pages / $1~$0.00975 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Mistral: Mistral Medium 3 | 64.9% | 923ms | ~684 pages / $1~$0.00146 per standard page | 80% | 100 | wide-model-selectionunified-apicost-effective |
| OpenAI: GPT-5.6 Luna | 64.8% | 2.4s | ~816 pages / $1~$0.00123 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Mistral: Mistral Medium 3.1 | 64.6% | 753ms | ~684 pages / $1~$0.00146 per standard page | 86.67% | 100 | wide-model-selectionunified-apicost-effective |
| Gemini 3 Flash Preview | 63.6% | 3.7s | ~184 pages / $1~$0.00543 per standard page | 100% | 100 | large-context-windowcost-effectivefast-inference |
| StepFun: Step 3.7 Flash | 63.6% | 5.1s | ~1,150 pages / $1~$0.000867 per standard page | 100% | — | large-context-windowcost-effective |
| Anthropic: Claude Opus 4.6 | 63.4% | 2.8s | ~46.1 pages / $1~$0.0217 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Claude Opus 4.5 | 63.1% | 5.3s | ~59.2 pages / $1~$0.0169 per standard page | 100% | 100 | |
| Qwen: Qwen3 VL 235B A22B Instruct | 63% | 27.1s | ~678 pages / $1~$0.00148 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Qwen3-VL 235B A22B Instruct (DeepInfra) | 63% | 11.5s | ~1,000 pages / $1~$0.000996 per standard page | 100% | 100 | fast-inferenceopen-weightscost-effective |
| Mistral Medium | 62.1% | 1.7s | ~739 pages / $1~$0.00135 per standard page | 100% | 100 | dedicated-ocr-modelsmultilingualeuropean-languages |
| Mistral Medium | 61.9% | 1.7s | ~602 pages / $1~$0.00166 per standard page | 100% | 100 | dedicated-ocr-modelsmultilingualeuropean-languages |
| Gemini 3.1 Flash Lite | 60.7% | 3.1s | ~821 pages / $1~$0.00122 per standard page | 98.29% | — | |
| Google: Gemini 3.1 Flash Lite | 60.7% | 1.6s | ~784 pages / $1~$0.00128 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.1 Flash Lite Preview | 60.5% | 971ms | ~845 pages / $1~$0.00118 per standard page | 100% | 100 | large-context-windowcost-effectivefast-inference |
| Mistral Large | 58.5% | 5.3s | ~477 pages / $1~$0.0021 per standard page | 100% | — | dedicated-ocr-modelsmultilingualeuropean-languages |
| Mistral: Mistral Large 3 2512 | 58.5% | 888ms | ~849 pages / $1~$0.00118 per standard page | 25% | 100 | wide-model-selectionunified-apicost-effective |
| Xiaomi: MiMo-V2.5 | 57.3% | 34.2s | ~2,620 pages / $1~$0.000381 per standard page | 93.33% | — | large-context-windowmillion-token-contextcost-effective |
| Google: Gemini 3.5 Flash-Lite | 56.9% | 1.1s | ~504 pages / $1~$0.00198 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.5 Flash Lite | 56.9% | 1.5s | ~544 pages / $1~$0.00184 per standard page | 80% | — | |
| xAI: Grok 4.5 | 56.5% | 5.9s | ~176 pages / $1~$0.00567 per standard page | 93.33% | — | native-pdf-supportlarge-context-window |
| Google: Gemini 3.6 Flash | 56.5% | 2.1s | ~189 pages / $1~$0.00529 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.6 Flash | 56.5% | 2.8s | ~179 pages / $1~$0.00559 per standard page | 80% | — | |
| Qwen: Qwen3 VL 30B A3B Instruct | 56.1% | 1.8s | ~1,940 pages / $1~$0.000517 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Qwen3-VL 30B A3B Instruct (DeepInfra) | 56.1% | 1.3s | ~2,470 pages / $1~$0.000404 per standard page | 100% | 100 | fast-inferenceopen-weightscost-effective |
| MiniMax: MiniMax M3 | 55% | 1.8s | ~940 pages / $1~$0.00106 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| xAI: Grok 4.3 | 54.3% | 3.7s | ~318 pages / $1~$0.00314 per standard page | 100% | — | large-context-windowmillion-token-context |
| Anthropic: Claude Sonnet 4.6 | 54.1% | 2.2s | ~76.9 pages / $1~$0.013 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| OpenAI: GPT-5.6 Luna Pro | 53.9% | 4.0s | ~189 pages / $1~$0.0053 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Mistral Small | 53.6% | 1.8s | ~4,560 pages / $1~$0.000219 per standard page | 100% | 100 | dedicated-ocr-modelsmultilingualeuropean-languages |
| Meta: Llama 4 Scout | 53.6% | 1.9s | ~1,370 pages / $1~$0.000731 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| ByteDance Seed: Seed-2.0-Code | 53% | 7.9s | ~259 pages / $1~$0.00385 per standard page | 100% | — | large-context-windowcost-effective |
| Qwen2.5-VL 32B Instruct (DeepInfra) | 51.6% | 1.1s | ~467 pages / $1$0.00214/page | 100% | 100 | fast-inferenceopen-weightscost-effective |
| Google: Gemini 3.7 Flash | 51.6% | 2.3s | ~226 pages / $1~$0.00442 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.7 Flash | 51.6% | 3.4s | ~217 pages / $1~$0.0046 per standard page | 80% | — | |
| Anthropic: Claude Fable 5 | 50.8% | 3.4s | ~27.3 pages / $1~$0.0367 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Anthropic: Claude Sonnet 4.5 | 50.3% | 3.1s | ~76.9 pages / $1~$0.013 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Anthropic: Claude Sonnet 4 | 49.5% | 1.7s | ~76.9 pages / $1~$0.013 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Meta: Muse Glimmer 30B | 49.2% | 1.7s | ~754 pages / $1~$0.00133 per standard page | 80% | — | large-context-windowcost-effective |
| Qwen: Qwen3.6 Flash | 49% | 10.4s | ~1,270 pages / $1~$0.000789 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Anthropic: Claude Sonnet 5 | 47.4% | 3.9s | ~136 pages / $1~$0.00733 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Qwen: Qwen3.8 27B | 47.3% | 2.5s | ~455 pages / $1~$0.0022 per standard page | 93.33% | — | large-context-windowcost-effective |
| GPT 4.1 Mini | 46.2% | 2.8s | ~412 pages / $1~$0.00243 per standard page | 100% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| Claude Opus 5 | 45.6% | 3.7s | ~45.2 pages / $1~$0.0221 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Qwen: Qwen3.6 27B | 45.1% | 1.8s | ~365 pages / $1~$0.00274 per standard page | 100% | — | large-context-windowcost-effective |
| xAI: Grok Build 0.1 | 45.1% | 8.7s | ~89.5 pages / $1~$0.0112 per standard page | 93.33% | — | large-context-window |
| Google: Gemini 3.5 Flash | 44.2% | 2.3s | ~59.6 pages / $1~$0.0168 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.5 Flash | 44.2% | 1.6s | ~51.3 pages / $1~$0.0195 per standard page | 100% | — | |
| Qwen: Qwen3.6 35B A3B | 43.8% | 1.8s | ~1,170 pages / $1~$0.000852 per standard page | 100% | — | large-context-windowcost-effective |
| OpenAI: GPT-4.1 Nano | 43.5% | 2.1s | ~3,040 pages / $1~$0.000329 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Thinking Machines: Inkling Small | 43.4% | 1.2s | ~389 pages / $1~$0.00257 per standard page | 80% | — | large-context-windowcost-effective |
| Claude Haiku 4.5 | 42.6% | 2.8s | ~300 pages / $1~$0.00333 per standard page | 100% | 100 | |
| Anthropic: Claude Haiku 4.5 | 42.6% | 1.7s | ~274 pages / $1~$0.00365 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| O4 Mini | 41.7% | 2.5s | ~72.1 pages / $1~$0.0139 per standard page | 100% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| Amazon: Nova 2 Lite | 41.6% | 1.1s | ~580 pages / $1~$0.00172 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Perceptron: Perceptron Mk1 | 41.2% | 4.1s | ~982 pages / $1~$0.00102 per standard page | 80% | — | cost-effective |
| SpaceXAI: Grok 4.6 | 41.1% | 12.1s | ~99.4 pages / $1~$0.0101 per standard page | 100% | — | native-pdf-supportlarge-context-window |
| Amazon: Nova Lite 1.0 | 40.1% | 3.2s | ~4,260 pages / $1~$0.000235 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| OpenAI: GPT-5.6 Terra Pro | 37.6% | 2.7s | ~50.7 pages / $1~$0.0197 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Perplexity: Sonar Pro | 37.1% | 5.2s | ~91.9 pages / $1~$0.0109 per standard page | 90% | 100 | wide-model-selectionunified-apicost-effective |
| GPT 4.1 | 36.1% | 2.0s | ~174 pages / $1~$0.00575 per standard page | 100% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| OpenAI: GPT-5.6 Sol | 36.1% | 2.3s | ~139 pages / $1~$0.00721 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-5.6 Sol Pro | 33.7% | 2.7s | ~77.3 pages / $1~$0.0129 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-5.1-Codex | 33.5% | 6.1s | ~135 pages / $1~$0.0074 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Perplexity: Sonar | 32.8% | 8.8s | ~966 pages / $1~$0.00104 per standard page | 80% | 100 | wide-model-selectionunified-apicost-effective |
| GPT 4o Mini | 32.3% | 1.2s | ~401 pages / $1~$0.00249 per standard page | 100% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| OpenAI: GPT-5.4 Image 2 | 31.6% | 1.6s | ~48.2 pages / $1~$0.0207 per standard page | 100% | — | native-pdf-supportlarge-context-window |
| Gemini 3.1 Pro Preview | 30.4% | 5.2s | ~21.7 pages / $1~$0.046 per standard page | 100% | 100 | large-context-windowcost-effectivefast-inference |
| Gemini 3.1 Pro Preview Custom Tools | 30.4% | 5.7s | ~18.3 pages / $1~$0.0546 per standard page | 100% | 100 | large-context-windowcost-effectivefast-inference |
| o3 | 30.4% | 2.3s | ~54.7 pages / $1~$0.0183 per standard page | 80% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| OpenAI: GPT-5.6 Terra | 29.5% | 1.8s | ~118 pages / $1~$0.00851 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-5.5 | 29.2% | 1.3s | ~35.6 pages / $1~$0.0281 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google: Gemini 2.5 Pro Preview 06-05 | 25.6% | 4.3s | ~41.3 pages / $1~$0.0242 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Gemini 2.5 Pro | 25.6% | 6.1s | ~25.1 pages / $1~$0.0398 per standard page | 100% | 100 | large-context-windowcost-effectivefast-inference |
| ByteDance: UI-TARS 7B | 21.5% | 973ms | ~4,180 pages / $1~$0.000239 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| Amazon: Nova Pro 1.0 | 21.1% | 1.3s | ~320 pages / $1~$0.00313 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| AWS Textract | 19.5% | 1.6s | ~623 pages / $1$0.001605/page | 100% | 30 | |
| docTR | 12.8% | 8.2s | 500 pages / $1$0.002/page | 100% | 15 | document layout analysistable detectionmodern deep learning architecture |
| EasyOCR | 4.8% | 3.7s | 500 pages / $1$0.002/page | 100% | 80 | excellent handwriting recognitiongood multilingual supportGPU accelerated |
| Mistral OCR 4.1 | No hard score | 4.3s | 200 pages / $1$0.005/page | 100% | — | |
| Qwen: Qwen3.7 Flash | No hard score | 11.9s | ~7,910 pages / $1~$0.000126 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Anthropic: Claude Haiku 5.5 | No hard score | 3.1s | ~2,200 pages / $1~$0.000455 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| NVIDIA: Nemotron 3.5 Content Safety | No hard score | 956ms | ~3,230 pages / $1~$0.000309 per standard page | 100% | — | large-context-windowcost-effective |
| Qwen: Qwen3.8 Omni Flash | No hard score | 5.9s | ~2,050 pages / $1~$0.000487 per standard page | 98.81% | — | large-context-windowmillion-token-contextcost-effective |
| DeepSeek: DeepSeek V4 Flash Vision Exp | No hard score | 4.5s | ~1,270 pages / $1~$0.000789 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Z.ai: GLM 5.3 Flash | No hard score | 2.6s | ~1,420 pages / $1~$0.000703 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Perceptron: Perceptron Mk1.5 | No hard score | 1.4s | ~895 pages / $1~$0.00112 per standard page | 100% | — | cost-effective |
| inclusionAI: Ling 3.0 Flash VL | No hard score | 6.4s | ~3,330 pages / $1~$0.000301 per standard page | 100% | — | large-context-windowcost-effective |
| DeepSeek: DeepSeek V4.1 Flash | No hard score | 3.8s | ~852 pages / $1~$0.00117 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Z.ai: GLM 5.3 FlashX | No hard score | 3.7s | ~571 pages / $1~$0.00175 per standard page | 81.12% | — | large-context-windowmillion-token-contextcost-effective |
| PrismML: Ternary Bonsai 2 27B | No hard score | 11.1s | ~2,240 pages / $1~$0.000447 per standard page | 92.86% | — | large-context-windowcost-effective |
| Llama 3.2 11B Vision (DeepInfra) | No hard score | 8.0s | ~1,170 pages / $1$0.000856/page | 99.17% | 100 | fast-inferenceopen-weightscost-effective |
| OpenAI: GPT-6 Luna Pro | No hard score | 10.4s | ~393 pages / $1~$0.00254 per standard page | 83.33% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6 Luna | No hard score | 6.5s | ~1,130 pages / $1~$0.000881 per standard page | 80.44% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Mistral: Mistral Small 3.1 24B | No hard score | 18.3s | ~1,510 pages / $1~$0.000663 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| DeepSeek: DeepSeek Flash Latest | No hard score | 10.9s | ~2,480 pages / $1~$0.000404 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Xiaomi: MiMo-V2.6-Flash | No hard score | 17.9s | ~3,310 pages / $1~$0.000302 per standard page | 91.67% | — | large-context-windowmillion-token-contextcost-effective |
| Pareto 26.9 | No hard score | 2.2s | ~142 pages / $1~$0.00702 per standard page | 100% | — | large-context-window |
| ScaleDown: Summarize | No hard score | 1.3s | ~28,900 pages / $1~$0.0000346 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Nano Banana 2.1 | No hard score | 1.7s | ~175 pages / $1~$0.0057 per standard page | 95.45% | — | lightweight/compact model sizemulti-language text recognitionscene text and natural image OCR |
| Meta: Muse Wonder | No hard score | 5.2s | ~60.9 pages / $1~$0.0164 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google: Gemini 4 Argon | No hard score | 3.4s | ~122 pages / $1~$0.00821 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 4 Argon | No hard score | 10.2s | ~132 pages / $1~$0.0076 per standard page | 91.67% | — | strong handwriting recognitionmulti-language and script supportscene text extraction in natural images |
| Mistral Large 4 | No hard score | 30.5s | ~556 pages / $1~$0.0018 per standard page | 77.78% | — | strong multilingual supporthandwriting recognitioncomplex layout analysis |
| Xiaomi: MiMo-V2.6-Pro | No hard score | 21.8s | ~1,060 pages / $1~$0.00094 per standard page | 91.03% | — | large-context-windowmillion-token-contextcost-effective |
| Cohere: Command A+ | No hard score | 4.4s | ~812 pages / $1~$0.00123 per standard page | 68.67% | — | large-context-windowcost-effective |
| StepFun: Step 5 Preview | No hard score | 3.6s | ~383 pages / $1~$0.00261 per standard page | 92.86% | — | large-context-windowmillion-token-context |
| IBM Docling (hosted) | No hard score | 6.6s | 250 pages / $1$0.004/page | 75% | — | |
| Mistral: Mistral Medium 3.5 | No hard score | 861ms | ~182 pages / $1~$0.0055 per standard page | 80% | — | large-context-window |
| Anthropic: Claude Sonnet 5.5 | No hard score | 2.6s | ~122 pages / $1~$0.00821 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Tesseract 5 | No hard score | 2.3s | 500 pages / $1$0.002/page | 100% | 116 | fastlightweightgood for printed text |
| Nex AGI: Nex-N2.5-Mini | No hard score | 35.5s | ~11,600 pages / $1~$0.0000864 per standard page | 62.86% | — | large-context-windowcost-effective |
| Z.ai: GLM Flash Latest | No hard score | 2.7s | ~12,300 pages / $1~$0.0000815 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Anthropic Claude Sonnet Latest | No hard score | 2.0s | ~136 pages / $1~$0.00733 per standard page | 100% | — | large-context-windowmillion-token-context |
| Gemini 2.5 Flash-Lite | No hard score | 3.7s | ~3,710 pages / $1~$0.00027 per standard page | 88.89% | 100 | large-context-windowcost-effectivefast-inference |
| Anthropic: Claude Fable 5.1 | No hard score | 4.0s | ~24.4 pages / $1~$0.041 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Mistral Medium 3 5 | No hard score | 882ms | ~166 pages / $1~$0.00603 per standard page | 100% | — | |
| Anthropic: Claude Opus 4.8 | No hard score | 1.9s | ~54.6 pages / $1~$0.0183 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| ByteDance Seed: Seed 2.1 Turbo | No hard score | 3.2s | ~320 pages / $1~$0.00313 per standard page | 100% | — | large-context-windowcost-effective |
| OCRmyPDF | No hard score | 1.5s | 500 pages / $1$0.002/page | 100% | 100 | PDF optimizationpreserves original PDF structureadds searchable text layer |
| Qwen: Qwen3.7 Plus | No hard score | 2.1s | ~731 pages / $1~$0.00137 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| Mistral: Mistral Large 4 | No hard score | 5.8s | ~515 pages / $1~$0.00194 per standard page | 100% | — | large-context-windowcost-effective |
| Pareto 26.10 Preview | No hard score | 2.1s | ~362 pages / $1~$0.00276 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| SpaceXAI: Grok 4.7 | No hard score | 4.7s | ~178 pages / $1~$0.00562 per standard page | 100% | — | native-pdf-supportlarge-context-window |
| RapidOCR | No hard score | 1.5s | 500 pages / $1$0.002/page | 100% | 80 | fast CPU inferencelightweightgood accuracy |
| Thinking Machines: Inkling | No hard score | 1.2s | ~240 pages / $1~$0.00416 per standard page | 100% | — | large-context-windowmillion-token-context |
| Qwen: Qwen3.5 Plus 2026-04-20 | No hard score | 1.9s | ~415 pages / $1~$0.00241 per standard page | 100% | — | large-context-windowmillion-token-contextcost-effective |
| MoonshotAI: Kimi K2.6 | No hard score | 10.3s | ~230 pages / $1~$0.00436 per standard page | 100% | — | large-context-windowcost-effective |
| Anthropic Claude Haiku Latest | No hard score | 1.7s | ~231 pages / $1~$0.00433 per standard page | 100% | — | large-context-window |
| Nex AGI: Nex-N2.5-Pro | No hard score | 26.7s | ~4,410 pages / $1~$0.000227 per standard page | 62.22% | — | large-context-windowcost-effective |
| Anthropic: Claude Opus Latest | No hard score | 3.8s | ~68.2 pages / $1~$0.0147 per standard page | 93.33% | — | large-context-windowmillion-token-context |
| Sakana: Fugu Max | No hard score | 7.6s | ~178 pages / $1~$0.00562 per standard page | 93.33% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Xiaomi: MiMo-V2.6-Pro-UltraSpeed | No hard score | 39.8s | ~106 pages / $1~$0.0094 per standard page | 93.33% | — | large-context-windowmillion-token-context |
| Anthropic: Claude Opus 5.5 | No hard score | 5.8s | ~60.9 pages / $1~$0.0164 per standard page | 90% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| MoonshotAI: Kimi K2.7 Code | No hard score | 2.9s | ~288 pages / $1~$0.00347 per standard page | 90% | — | large-context-windowcost-effective |
| MoonshotAI Kimi Latest | No hard score | 2.4s | ~68.5 pages / $1~$0.0146 per standard page | 86.67% | — | large-context-windowcost-effective |
| Kimi K3 (Fireworks) | No hard score | 5.0s | ~24.1 pages / $1~$0.0414 per standard page | 86.67% | — | |
| Qwen: Qwen3.8 Max (0902) | No hard score | 5.5s | ~178 pages / $1~$0.00562 per standard page | 86.67% | — | large-context-windowmillion-token-context |
| Qwen: Qwen3.8 Flash | No hard score | 6.3s | ~2,200 pages / $1~$0.000454 per standard page | 88.89% | — | large-context-windowmillion-token-contextcost-effective |
| OpenAI: GPT-5.5 Pro | No hard score | 7.0s | ~7.88 pages / $1~$0.127 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT Chat Latest | No hard score | 1.9s | ~47.6 pages / $1~$0.021 per standard page | 80% | — | native-pdf-supportlarge-context-window |
| Qwen: Qwen3.8 Max Prime | No hard score | 4.3s | ~89 pages / $1~$0.0112 per standard page | 77.14% | — | large-context-windowmillion-token-context |
| Anthropic: Claude Fable Latest | No hard score | 5.0s | ~27.3 pages / $1~$0.0367 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6 Sol Pro | No hard score | 3.8s | ~122 pages / $1~$0.00821 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6.1 Sol | No hard score | 3.5s | ~122 pages / $1~$0.00821 per standard page | 75% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Sakana: Fugu Ultra v2 | No hard score | 14.2s | ~42.1 pages / $1~$0.0238 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI GPT Sol Latest | No hard score | 2.3s | ~122 pages / $1~$0.00821 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google: Nano Banana 2.1 | No hard score | 1.7s | ~162 pages / $1~$0.00616 per standard page | 66.67% | — | |
| OpenAI GPT Terra Latest | No hard score | 1.7s | ~105 pages / $1~$0.0095 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6 Sol | No hard score | 2.2s | ~122 pages / $1~$0.00821 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google: Gemini 3.8 Flash | No hard score | 10.3s | ~325 pages / $1~$0.00308 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Qwen: Qwen3.8 Max | No hard score | 5.6s | ~157 pages / $1~$0.00636 per standard page | 80% | — | large-context-windowmillion-token-context |
| Meta: Muse Spark 1.1 | No hard score | 2.2s | ~143 pages / $1~$0.007 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google Gemini Flash Latest | No hard score | 3.3s | ~166 pages / $1~$0.00601 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Google Gemini Pro Latest | No hard score | 5.1s | ~27.6 pages / $1~$0.0362 per standard page | 100% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6 Astra | No hard score | 3.1s | ~24.4 pages / $1~$0.041 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6.1 Sol Pro | No hard score | 14.1s | ~122 pages / $1~$0.00821 per standard page | 75% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Sakana: Fugu Ultra | No hard score | 16.3s | ~13.4 pages / $1~$0.0748 per standard page | 100% | — | large-context-windowmillion-token-context |
| OpenAI GPT Luna Latest | No hard score | 2.7s | ~2,440 pages / $1~$0.00041 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| OpenAI: GPT-6 Astra Pro | No hard score | 4.7s | ~24.4 pages / $1~$0.041 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| MoonshotAI: Kimi K3 | No hard score | 3.6s | ~75.2 pages / $1~$0.0133 per standard page | 80% | — | large-context-windowmillion-token-context |
| OpenAI GPT Mini Latest | No hard score | 1.4s | ~313 pages / $1~$0.00319 per standard page | 80% | — | native-pdf-supportlarge-context-windowcost-effective |
| Meta: Muse Spark 1.3 | No hard score | 5.3s | ~261 pages / $1~$0.00383 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Gemini 3.8 Flash | No hard score | 3.1s | ~351 pages / $1~$0.00285 per standard page | 83.33% | — | |
| o1 | No hard score | 5.2s | ~10.8 pages / $1~$0.0922 per standard page | 100% | 100 | high-accuracyfast-inferencestructured-document-extraction |
| Anthropic: Claude Opus 4.1 | No hard score | 7.9s | ~18.3 pages / $1~$0.0547 per standard page | 100% | 100 | wide-model-selectionunified-apicost-effective |
| OpenAI GPT Astra Latest | No hard score | 2.7s | ~24.4 pages / $1~$0.041 per standard page | 80% | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Fireworks: Ember-1 | No hard score | 4.1s | ~81.2 pages / $1~$0.0123 per standard page | 80% | — | large-context-windowmillion-token-context |
| ScaleDown: Extract | No hard score | 190ms | ~28,900 pages / $1~$0.0000346 per standard page | — | — | native-pdf-supportlarge-context-windowmillion-token-context |
| Pareto Latest | No hard score | — | ~142 pages / $1~$0.00702 per standard page | — | — | large-context-window |
| Amazon: Nova Premier 1.0 | No hard score | 384ms | ~87 pages / $1~$0.0115 per standard page | — | 100 | wide-model-selectionunified-apicost-effective |
OpenOCR sends each document to the selected engine. That engine may run on OpenOCR-operated infrastructure or an external provider.
Provider routing
The selected provider receives the document content needed for OCR
Request records
Persisted requests include metadata, submitted URLs, and extracted text
Provider policies
Retention, training, moderation, and data regions vary by provider
Read the full privacy and data-retention policy.
Want to test engines on your own documents? Try side-by-side comparison → · How we benchmark →