rankings / read-documents
The best models for reading documents
18 models & services · 1 callable here now
Ranked by benchmark score per dollar (quality floor applied). Scores: olmOCR-bench — how accurately a model reads and parses real documents. Fetched 2026-07-12. Prices are our live per-call rates; ~ marks an estimate until the model is onboarded.
sorted by value · sort by score
| # | model | score | price | params | status |
|---|---|---|---|---|---|
| 1 | deepseek-ai/DeepSeek-OCR best value | 75.7 (#13) | $0.1667 / 1M tokens | 3336.1M | live · use now |
| 2 | infly/Infinity-Parser2-Pro top score | 87.6 (#1) | - | 35107.2M | not hosted |
| 3 | datalab-to/chandra-ocr-2 | 85.8 (#2) | - | 5295.6M | not hosted |
| 4 | rednote-hilab/dots.mocr | 83.9 (#3) | - | 3039.2M | not hosted |
| 5 | datalab-to/surya-ocr-2 | 83.3 (#4) | - | 686.2M | not hosted |
| 6 | lightonai/LightOnOCR-2-1B | 83.2 (#5) | - | 1005.6M | not hosted |
| 7 | datalab-to/chandra | 83.1 (#6) | - | 8767.1M | not hosted |
| 8 | infly/Infinity-Parser-7B | 82.5 (#7) | - | 8292.2M | not hosted |
| 9 | tiiuae/Falcon-OCR | 80.3 (#8) | - | 269.9M | not hosted |
| 10 | baidu/Qianfan-OCR | 79.8 (#9) | - | 4741.4M | not hosted |
| 11 | rednote-hilab/dots.ocr | 79.1 (#10) | - | 3039.2M | not hosted |
| 12 | deepseek-ai/DeepSeek-OCR-2 | 76.3 (#11) | - | 3389.1M | not hosted |
| 13 | lightonai/LightOnOCR-1B-1025 | 76.1 (#12) | - | 1161.2M | not hosted |
| 14 | opendatalab/MinerU2.5-2509-1.2B | 75.2 (#14) | - | 1156M | not hosted |
| 15 | zai-org/GLM-OCR | 75.2 (#15) | - | 1325.3M | not hosted |
| 16 | Mistral OCR | 72src (#16) | $1.00 / 1k pages | - | reference |
| 17 | OpenAI GPT-4o | 69.9src (#17) | per-token vision pricing | - | reference |
| 18 | Google Gemini Flash 2 | 63.8src (#18) | per-token vision pricing | - | reference |