Receipt Scanner
A receipt photo becomes fields you can total, file and audit — one credit each.
VisionParse reads a receipt into structured fields — merchant, receipt number, date, time, every line with its quantity and amount, subtotal, tax, total and how it was paid.
We measured it the way receipts actually arrive: the same thermal slip clean, faded, creased, photographed at an angle under uneven light, and shrunk to a 600-pixel phone crop. All fourteen fields came back correct in every condition, and on an itemised A4 receipt with a unit-price column all eighteen did. The arithmetic is checked — lines summed against the subtotal and the total before you see them.
- AI tier · 1 credit a document
- JPG · PNG · WEBP · PDF
- Photos auto-compressed · PDF to 25 pages
- Deleted within 1h
What this tool does, in numbers.
- ACCURACY
All fourteen fields — merchant, receipt number, date, time, currency, five lines, subtotal, tax, total and payment — read exactly on every condition tested.
- LIMIT
One field has ever gone wrong: a date written 02/08/2026, read month-first in two runs out of thirty-three.
- SPEED
A receipt returns as structured fields in about 1.8 seconds, arithmetic checks included.
- COST
One AI credit a receipt, whether it carries five lines or a full weekly shop.
- ACCESS
Needs an account with AI credits: every paid plan includes them monthly, or a one-time pack adds them without a subscription.
- RETENTION
The uploaded receipt is deleted within an hour, the extracted fields within 30 days, and nothing trains a model.
Where the 14 of 14 fields, five conditions comes from.
Every figure on this page traces to a dated run on hardware we name. Nothing here is a vendor estimate.
Measured 2026-08-04
- Claim
- Receipts survive the state they arrive in: faded, creased, photographed at an angle or shrunk to a phone crop, every field still read exactly.
- Method
- AI extraction tier (gemini-3.5-flash-lite), receipt schema, one credit. Full methodology.
- Sample
- A thermal-layout supermarket receipt (five lines, VAT line, card payment) rendered five ways — clean 200 dpi, faded contrast, creased, photographed at an angle under uneven light at JPEG quality 75, and a 600-pixel-wide crop — then a ten-rung ladder down to 200 pixels wide, JPEG quality 10 and heavy blur. Separately, an itemised A4 sales receipt with a unit-price column, a named customer and a 20% VAT line: 18 of 18.
- Measured
2026-08-04on Contabo Cloud VPS 8 — 8 vCPU AMD EPYC, 24 GB RAM, Ubuntu 24.04.4- Limitations
- Both are our own fixtures, not slips from a real till, and neither carries a tip — that field is supported but unmeasured. Non-English receipts are unmeasured too. Send real receipts and we measure them before you rely on it.
- Contact
- VisionParse@senithu.lk
What this tool gets wrong
Degradation barely moved it. Ten rungs down — 200 pixels wide, JPEG quality 10, heavy blur, faded thermal print, a fold through the middle — returned all thirteen fields correctly on every rung but one. Receipts are short, high-contrast and predictable in layout, which is why they hold up where dense documents would not.
A date written 02/08/2026 is genuinely ambiguous, and no engine can settle it from pixels. Day-first it is 2 August; month-first it is 8 February. Across 33 runs it was read correctly 31 times, and the two misses landed on different rungs of the degradation ladder — which is what ambiguity looks like, rather than a resolution limit. The model itself listed the date as low-confidence in six of eight repeat runs on the identical image, and that flag is shown to you rather than hidden: a silently wrong date is the error that costs an accounts team real money.
The unit price is read, never derived — so the arithmetic check means something. A till slip prints a quantity and a line total and no unit price. Asked for one anyway, a model will divide — and then our "quantity times unit price equals amount" check is grading that same division and cannot fail. The schema now says printed-or-null, so on an itemised receipt the check runs against a real second figure, and on a till slip it stays quiet instead of reporting verified arithmetic that was never verified.
Tips are supported but unmeasured, and both fixtures are ours. Non-zero tax is now measured — a 20% VAT line of £70.80 read exactly on an itemised A4 receipt — but neither fixture carries a restaurant tip, so we will not claim a figure for one. Nor are these slips from a real till. Real receipts are welcome: measuring them first is how every figure on this site got here.
Three steps, no account.
Drop the receipt
A photo from your phone or a scan — JPG, PNG, WebP or PDF. Large photos are compressed in the browser before upload.
The AI engine reads it
One credit a receipt, about 1.8 seconds. Every line, the totals and the payment method come back as named fields.
Check the arithmetic, take the JSON
The lines are summed against the subtotal and total, and anything the model doubted is flagged. Copy the JSON or download it.
What this tool accepts.
| Property | Free tool |
|---|---|
| Engine | AI extraction tier, 1 credit per receipt |
| Measured result | 14 of 14 fields across five conditions; 18 of 18 on an itemised A4 receipt (2026-08-04) |
| Hardest rung passed | 200 pixels wide, JPEG quality 10, heavy blur |
| Fields returned | Merchant, address, receipt number, date, time, customer, currency, items (description, quantity, unit price, amount), subtotal, tax, tip, total, payment method |
| Checks | Line items summed against subtotal; subtotal plus tax against total; quantity times unit price where a unit price is printed |
| Input formats | JPG, PNG, WEBP, PDF — up to 25 pages a credit |
| Access | Account with AI credits — any paid plan, or a one-time pack |
| File retention | Originals deleted within 1 hour; results within 30 days |
When to use the API instead.
This page converts one file now. To convert files from your own software, the same engine runs over HTTP from Starter upward.
# same engine, from your code curl -X POST https://api.visionparse.app/v1/ocr \ -H "x-api-key: $VISIONPARSE_KEY" \ -F "file=@image.jpg" → 202 { "jobId": "c0df8e2c…" }
Use the API when
You convert more than a few files, convert on a schedule, or need results inside another system. Batch, webhooks and bounding boxes are API-only.
Frequently asked
Does a crumpled photo of a faded receipt actually work?
That is what we measured. The same slip was tested faded, creased, photographed at an angle under uneven light, and shrunk to a 600-pixel crop — fourteen of fourteen fields came back correct in every one of those conditions, and it kept working down to 200 pixels wide and JPEG quality 10.
Will it get the date right?
Usually, and it tells you when it is unsure. A receipt printed 02/08/2026 could be 2 August or 8 February — nothing in the image settles it. In 33 runs it read correctly 31 times, and it flagged the date as low-confidence in six of eight repeat runs on the identical image. That flag appears with your result.
Why is this not free like your other tools?
Because reading fields is a different job from reading text. Free OCR gives you the words; turning them into merchant, date, line items and totals needs the AI engine, which costs money per call. So it runs on credits — included monthly on every paid plan, or bought once in a pack.
Can I scan a stack of receipts for expenses?
One at a time in the browser today, one credit each, and through the API if you want to automate it. Batch upload is on the roadmap and is not built yet — we do not list features that do not exist.
What about restaurant tips and VAT?
VAT is measured: a 20% line of £70.80 read exactly on an itemised A4 receipt. Tips are not — neither fixture carries one, so the field exists and comes back populated but we will not claim a figure for it. Send a few real slips and we will measure them properly.