Typhoon-OCR-7b Now Available!!

Typhoon-OCR-7b is now available through Float16's AI as a Service today!

Typhoon-OCR-7b Details

Typhoon-OCR-7b is a Model from the Typhoon team (SCB10X), built on top of Qwen-2.5-vl-7b Model.

Typhoon-OCR-7b has better OCR performance than GPT-4o and Gemini 2.5.

It can be used for various applications such as:

• Document classification
• Reading various cards
• Reading invoices or receipts

---

How to Use

Prerequisites:

1. API key
2. typhoon-ocr Library

API Key

API key can be obtained by registering at Float16-App

Then go to AI as a Service -> API Key

Typhoon-ocr

Typhoon-ocr can be installed via pip using command pip install typhoon-ocr

from typhoon_ocr import ocr_document

# Process PDF document
markdown = ocr_document(
    pdf_or_image_path="document.pdf",  # Works with PDFs or images
    task_type="default",               # Choose between "default" or "structure"
    page_num=2,                        # Process page 2 of a PDF (default is 1, always 1 for images)
    #Config Base URL and API Key
    base_url="https://api.float16.cloud/v1",
    api_key="float16-sk-xxxxxxx"
)

How to use Typhoon-OCR through Float16-API

Change base_url and api_key to Float16's to start using.

More details about Typhoon-OCR

Performance

Usage	Duration
Sequential	Parallel
67 seconds	20 seconds

Duration for processing 6-page document

Sequential usage means waiting for previous processing to complete before processing next.

Parallel usage means sending all jobs for processing simultaneously.

from concurrent.futures import ThreadPoolExecutor
def process_page(page_num):
    return ocr_document(
        pdf_or_image_path="document.pdf",
        task_type="default",
        page_num=page_num,
        base_url="https://api.float16.cloud/v1",
        api_key="float16-sk-xxxxxxx"
    )
start_time = time.time()
with ThreadPoolExecutor(max_workers=6) as executor:
    futures = [executor.submit(process_page, i) for i in range(6)]
    results = [future.result() for future in futures]
print("Time taken for parallel processing of 6 pages:", time.time() - start_time)

Rate-Limit

• 10 Requests/sec
• First-Come, First-Serve

Speed per Page

Typhoon-OCR-7b provides varying speed per page depending on the amount of text in the document. More text means longer processing time.

Average time for processing is 40 - 60 words per second per Request.

Pricing

$0.03 per page or approximately 1 baht per page

Trial

You can receive free Credit of $5.00 daily (resets daily at 07:00), which allows approximately 150 pages of free OCR per day.

Payment

Payment can be made through Stripe with any type of credit card, or contact Float16 directly for payment through all communication channels.

Contact

• Medium : Float16.cloud
• Facebook : Float16.cloud
• X : Float16.cloud
• Discord : Float16.cloud
• Youtube : Float16.cloud
• Email : business[at]float16.cloud