One image, two profiles: read it or understand it
A live demo of two gateway profiles on the same image: ocr extracts the text as it stands, vision describes what is there and what follows.
Your image
POST /api/v1/demo/foto
Do not upload documents with personal data.
The image is sent to an external model provider; the answer stays in our request log for 7 days — we need it to investigate failures. How this works — in the privacy policy.
A demo with a daily cap: it exists to be tried, not to replace a service.
The answer appears here. Run the same image through both profiles — the difference between reading and understanding shows up best that way.
The hard case: when the pricier profile pays for itself
service: vision · vision-hardOne task, two vision profiles. A stock sheet in which one row's quantity times price does not match the printed line total. The truth is known in advance, so the answer can be checked rather than admired.
What is planted: the row "Door hinge", 14 times 87.50 equals 1225.00, printed as 1125.00. Every other row adds up.
Pick a profile and compare the answers. Exactly one error is planted.
Our measurement, 10 Aug 2026, four runs per profile, same question. On this sheet both profiles found the planted error 4 times out of 4 - the ordinary one in 1.3-2.1 s, the hard one in 6.3-7.8 s. On the Russian twin of the same sheet the picture was different: ordinary vision found it once out of four and three times declared "no discrepancies", while the hard profile scored 4 out of 4. We show both results rather than the flattering one: which profile earns its price is decided by a measurement on your own data, not by our table.