Truly serverless
Models stay warm so you skip cold starts. You pay per page, never for idle GPUs.
Turn any document into clean, structured text.
Oh See Arr runs state-of-the-art open OCR models on demand — no GPUs to manage, no cold starts to babysit. Upload a page, pick a model, and get back faithful Markdown, tables, and layout in milliseconds.
- $0.001
- per page
- 100+
- languages
- $0
- idle cost
Models we support
Open models, one serverless API.
PaddleOCR-VL and DeepSeek-OCR-2 are live at $0.001 per page. More models are on the way — switch anytime with a single parameter.
PaddlePaddle
PaddleOCR-VL 1.6
A compact multilingual VLM that parses text, tables, formulas, and handwriting across 100+ languages.
$0.001 / page~0.9B parametersApache 2.0
View model
DeepSeek
DeepSeek-OCR-2
Optical context compression that packs entire pages into a handful of vision tokens.
$0.001 / page~3B (MoE) parametersMIT
View model
OpenDataLab
MinerU 2.5 Pro
A 1.2B document-to-Markdown engine purpose-built for scientific PDFs, formulas, and tables.
1.2B parametersAGPL-3.0
Coming soon
LightOn
LightOnOCR-2-1B
A 1B end-to-end OCR VLM tuned for blazing throughput at production scale.
1B parametersApache 2.0
Coming soonWhy Oh See Arr
Production OCR without the ops.
Built for scale
Autoscaling batched inference handles one page or a million without config.
Structured output
Get Markdown, tables, and layout — ready to drop into RAG and data pipelines.
Private by default
Documents are processed in isolation and never used to train any model.