HINOW model
hinow/hivision
HiVision
Dedicated image reading: scanned documents, screenshots, photographs and charts.
Best for: Documentos, prints, gráficos e fotos
Pricing per 1 million tokens
Input
US$ 1,00
per 1 million tokens
Output
US$ 3,00
per 1 million tokens
- Scanned documents — invoices, contracts, forms and receipts that arrive as images.
- Screenshots — support tickets where the user sends a screenshot instead of describing the error.
- Charts and dashboards — reading what a chart shows and returning it as text or JSON.
- Photographs — product inspection, equipment condition, a photo sent by a customer.
- Structured extraction — turning any of these images into fields ready for your database.
HINOW's language models also accept images. HiVision stands out on two points that matter once volume grows:
- It is the fastest at visual tasks — on the same image, it answered in a few seconds, while general-purpose models took considerably longer.
- It charges less for the image — the same image consumes fewer input tokens here than on general-purpose models, and the input price is lower than HiMax's.
The side-by-side numbers are in Image understanding.
- 1
Send the image in the message
The
usercontent becomes a list with one text part and oneimage_urlpart. The image can be a public URL or adata:URL with base64. - 2
Ask for the output format
State the language of the answer and, when it feeds a system, ask for JSON only with the fields you expect.
- 3
Validate before storing
Treat the output as untrusted text: check types, required fields and ranges before writing to the database.
curl https://api.hinow.ai/v1/chat/completions \
-H "Authorization: Bearer $HINOW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hinow/hivision",
"messages": [{
"role": "user",
"content": [
{"type": "text", "text": "Extraia numero, CNPJ, total e vencimento. Responda só JSON."},
{"type": "image_url", "image_url": {"url": "https://exemplo.com/nota.png"}}
]
}]
}'Ask for the language and the format
Without an explicit instruction, the answer may come in English and open with the model introducing itself. Asking for "reply with JSON only" brings the output back clean — that is how it should be used inside a system.
See the full guide
The two ways to send an image, accepted formats, cost per image, model comparison and the errors that show up in production.