EN ▾

Uncensored LLM for OCR post-processinghttps://api.llmocrapi.com/v1

llmocrapi.comGuide

OCR API Pricing: What Text Cleanup Really Costs

OCR API pricing is normally quoted per page. Here you pay per token for cleaning the text your OCR engine already produced: $0.15 per 1M input tokens and $0.50 per 1M output tokens, billed from prepaid credit.

Updated

Isometric illustration of paper sheets turning into tokens and coins, representing per-token OCR cleanup pricing

Key points

  • The API reads text, not images, so you pay only for the cleanup step after your OCR engine runs.
  • A typical dense page costs a fraction of a cent: about $0.55 per 1,000 pages in the worked example below.
  • Output tokens cost more than input tokens, so asking for corrections only is the cheapest way to lower the bill.
  • Credit never expires, failed or refused requests are not charged, and new users can claim $0.50 of trial credit.

What you are paying for

This service does not read pixels. It does not accept scans, photos or PDFs as images, and it has no image input at all. What it accepts is plain text, which makes it the second stage of an OCR pipeline: your OCR engine turns an image into raw text, and the language model repairs that text. It rejoins hyphenated words, fixes rn that should be m, restores paragraph breaks, removes running headers and puts tables back into a readable shape.

That split matters for pricing. A per-page OCR quote bundles the recognition work and the cleanup into one number. Here the bill covers only the language model call, and it is metered in tokens, the same unit used by every OpenAI-compatible endpoint. If you already pay for an OCR engine, or run an open-source one on your own hardware, you add this cost on top of it. If you do not have an OCR engine at all, this API alone will not replace one.

The model behind the endpoint is one open-weight model, available under the ID uncensored. It is tuned to answer without content refusals for lawful adult use, which is why it suits archives that contain difficult material that other services reject or flag. For the exact cleanup workflow, see the documentation.

The two rates

There are two numbers to remember, and nothing else is added to them. No seat fee, no monthly minimum, no charge per page.

Token typePrice per 1M tokensPrice per 1,000 tokens
Input (prompt: instructions plus raw OCR text)$0.15$0.00015
Output (the cleaned text the model writes back)$0.50$0.0005

Input tokens are everything you send: your system prompt, any examples, and the OCR text itself. Output tokens are everything the model returns. The context window is 100,000 tokens for prompt and completion together, so a single request can hold a very long document, but it is almost always cheaper and more accurate to clean in page-sized or chapter-sized pieces.

Failed requests and refused requests are not charged. If a call returns an error, you pay nothing for it, which makes retry logic cheap to build.

Worked example: 1,000 pages

Take a batch of 1,000 scanned pages. Assume a dense page of roughly 600 words, which is about 800 tokens of OCR text. Assume the cleaned page comes back at about the same length, 800 tokens, and that your cleanup instructions add 200 tokens to every request.

  • Input per page: 800 + 200 = 1,000 tokens. For 1,000 pages that is 1,000,000 tokens, or 1M. At $0.15 per 1M, input costs $0.15.
  • Output per page: 800 tokens. For 1,000 pages that is 800,000 tokens, or 0.8M. At $0.50 per 1M, output costs $0.40.
  • Total: $0.15 + $0.40 = $0.55 for 1,000 pages.

Those are assumptions, not measurements of your documents. Token counts depend on language, formatting and how noisy the scan is. Text in languages with long compound words or non-Latin scripts usually uses more tokens per word, so measure a sample of your own pages before you budget a large job. The usage field in every response tells you the exact input and output tokens, so a 20-page pilot gives you a reliable per-page figure.

Scaling the same job

Because there is no tier structure, cost scales in a straight line with the volume. Using the same 1,000-token input and 800-token output per page:

PagesInput costOutput costTotal
1,000$0.15$0.40$0.55
10,000$1.50$4.00$5.50
100,000$15.00$40.00$55.00
1,000,000$150.00$400.00$550.00

The smallest top-up is $20, which covers roughly 36,000 pages at this profile ($20 divided by $0.55 per 1,000 pages is about 36.4 thousand). A large archive of a million pages is a $550 job under these assumptions, and you can fund it in steps rather than all at once.

What moves the bill

Output length

Output tokens cost more than three times as much as input tokens, so the biggest saving is to make the model write less. Many cleanup tasks do not need the whole page back. If you ask for only the corrected lines, or for a list of find-and-replace pairs, a page that previously produced 800 output tokens may produce 200. Rerun the example: input stays at $0.15 per 1,000 pages, output becomes 200,000 tokens at $0.50 per 1M, which is $0.10, so the total drops from $0.55 to $0.25. The trade-off is a little more code on your side to apply the changes.

Instruction length

A 200-token system prompt is cheap, but a 2,000-token prompt with ten examples is paid again on every request. At 1,000 pages that extra 1,800 tokens per page adds 1.8M input tokens, or $0.27. It can be worth it if the examples raise accuracy; just know that it is not free. Keep examples short, and drop ones that do not change the output.

Chunking

Splitting a long document into pieces repeats the instructions for each piece, but sending a whole book as one request makes errors harder to isolate and retries more expensive. A page or a section per request is a sensible default. Set max_tokens only if you want a hard ceiling on output; it is optional.

Retries

Failed requests are not charged, so retrying after a 429 costs only the time. Streamed requests hold a concurrency slot until the stream ends, which matters for throughput more than for price.

Prepaid credit, bonuses and the trial

You pay in advance and spend the balance down. There is no subscription, and credit never expires, so an unused balance stays available for the next batch. Top up any whole amount from $20 to $500. Larger top-ups include a bonus:

Top-upBonusCredit you receive
$20 to $49nonesame as the amount paid
$50+5%$52.50
$100+10%$110.00
$200+15%$230.00
$500+15%$575.00

This site accepts crypto top-ups: USDT on TRC20 or USDC on Base. The exact steps are on the pricing page. Every response carries an X-Balance-USD header, so a batch script can watch the balance and stop before it runs dry.

To test the cleanup on your own pages first, new users can claim $0.50 of trial credit with one click on the API key page after signing up with Google or with email and password. It needs no payment, is limited to one per person and is valid for 7 days. At the 1,000-pages-for-$0.55 profile, $0.50 covers about 900 pages. A trial key allows 2 concurrent requests and 60 requests per minute, enough for a realistic sample.

Comparing this with per-page quotes

To compare fairly, convert any per-page OCR quote to a price per 1,000 pages, and then add the recognition engine you would pair with this cleanup step. Check each provider's current pricing yourself, since quotes change and often depend on page size, features and volume tiers.

The token model is a better fit when your pages vary a lot in length, when most of your cost is cleanup rather than recognition, or when you want to process odd inputs such as logs, receipts and forms that do not map cleanly to a page. A per-page model can be easier to forecast when every page is the same size, and if you want one vendor to handle both recognition and cleanup, a full document-AI service may be simpler than two tools. If your use case includes flagging sensitive images, read the cost notes on NSFW detection, which looks at cost and trade-offs for another text-only use.

Questions and answers

Does this API do the OCR itself?

No. It accepts text only, with no image, audio or video input. Run your own OCR engine first, then send the raw text for cleanup.

How much does it cost to clean one page?

With 1,000 input tokens and 800 output tokens, one page costs $0.00055, or $0.55 per 1,000 pages. Your own figure depends on page length and how much text you ask the model to return.

Is there a monthly fee or minimum?

No. You pay per token from prepaid credit. The smallest top-up is $20, credit never expires, and failed or refused requests are not charged.

Can I try it before paying?

Yes. New users can claim $0.50 of trial credit on the API key page after sign-up. It needs no payment and is valid for 7 days.

Your key is one form away

Create an account, copy the key, change the base URL. That is the whole setup.

Get API key