Thai Vehicle License Plate OCR

Reads a Thai licence plate from a photograph: the number, the province, and the vehicle's make, model, body type, colour and orientation.

Version 3.0 on the main endpoints

Since 3 October 2026 the main endpoints serve version 3.0. The …/license-plate-ocr-v3 paths remain available as aliases at the same price. Every existing field keeps its name and format, so no code changes are needed.

POST/v3/store/smart-city/license-plate-ocr

+2 endpoints

Try it

Price0.75 ICper request
Measured99.4%
Capacity11,000 images/hour
Served20.5Mcalls

Input

OutputPOST /v3/store/smart-city/license-plate-ocr

conf: 93.39
country: th
lp_number: ขก5337
province: th-83:Phuket (ภูเก็ต)
vehicle_body_type: pickup
vehicle_brand: isuzu
vehicle_color: red
vehicle_model: isuzu_d-max
vehicle_orientation: 315
vehicle_year: 2020-present
is_vehicle: yes
is_missing_plate: no
message: success
status: 200

Every field, under ResponseOther endpoints: /base64, /base64 see Request

Request

curl -X POST https://api.iapp.co.th/v3/store/smart-city/license-plate-ocr \
  -H "apikey: YOUR_API_KEY" \
  -F "file=@vehicle.jpg"

Response

200OK. The fields are explained under Response.

{
    "conf": 93.39,
    "country": "th",
    "lp_number": "ขก5337",
    "province": "th-83:Phuket (ภูเก็ต)",
    "vehicle_body_type": "pickup",
    "vehicle_brand": "isuzu",
    "vehicle_color": "red",
    "vehicle_model": "isuzu_d-max",
    "vehicle_orientation": "315",
    "vehicle_year": "2020-present",
    "is_vehicle": "yes",
    "is_missing_plate": "no",
    "message": "success",
    "status": 200
}

Request

multipart/form-data, or the image as Base64 to /base64; 0.75 IC per request on every endpoint.

Headers

apikeystringrequired

Your API key, created under API Keys in the control panel.

Body

filefilerequired

The photograph. JPEG or PNG, up to 2 MB. On /base64, send JSON {"image": "<base64>"} instead.

Endpoint	Engine
`/v3/store/smart-city/license-plate-ocr`, `…/base64`	3.0 since 3 October 2026 (2.0 before)
`/v3/store/smart-city/license-plate-ocr-v3`, `…/base64`	3.0, alias of the main endpoints

The legacy paths /iapp_license_plate_recognition_v1_file, …_v1_base64, /license-plate-recognition/file and …/base64 follow the main endpoints at the same price.

Response

As in the panel, from both engines:

lp_numberprovincecountry

The plate, the province as th-83:Phuket (ภูเก็ต), and the country: th, my, mm, la

confnumber

Confidence in the plate reading, in percent

vehicle_brandvehicle_modelvehicle_year

Lowercase slugs, isuzu and isuzu_d-max, and the model's year range

vehicle_body_typevehicle_colorvehicle_orientation

sedan, pickup, suv and so on; the colour; the angle in degrees

is_vehicleis_missing_plate

yes or no

messagestatus

success, and the HTTP status

Version 3.0 adds vehicle_make_model_slug and make_model_conf, 0 to 1; vehicle_make_model_topk, the top five candidates with probabilities; vehicle_logo_brand and logo_brand_conf, the brand read independently from the badge and grille; vehicle_fuel_type: EV, Hybrid, Plug-in Hybrid, Petrol, Diesel, or multiple when the model is sold with several, listed in vehicle_fuel_options; and country_support_level: production, beta or detection_only.

Details

Measured

853 held-out images of today's Thai traffic covering all 73 models, never used in training; both engines read the same images. Method and per-model results: the benchmark paper (PDF).

Measure	Previous engine, v2.0, 2023 catalogue	Version 3.0
Brand correct	38.5%	99.4%
Brand correct, models within the v2.0 catalogue	41.8%	99.3%
Make and model correct, top-1	10.4%	99.18%
Make and model within top-5	—	100.0%
Returns an answer	about 50%	100%
Median time	0.2 s per image	0.5 s per image
Sustained throughput	—	11,000 images per hour at concurrency 4, production, 28 Aug 2026

310 of the 853 images, 27 of the 73 models, are vehicles introduced after the 2.0 catalogue was built, which 2.0 cannot identify by design. Plate, province, colour and orientation reading are unchanged; 3.0 is deterministic, the same vehicle giving the same answer on every scan, tight grille close-ups from gate cameras included.

Limits

Neighbouring countries: Malaysia in production, Myanmar in beta, Laos detection only.
Fuel type comes from the identified model through a maintained table, never from the image.

Data handling

Images are processed in memory and not kept after the response; no image goes to an external service. The service is GDPR and PDPA compliant. For an on-premise deployment, in which no photograph leaves your network, contact us.

Changelog

Version	Date	Changes
3.0.0	3 Oct 2026	The main endpoints move from version 2.0 to 3.0; the `-v3` paths remain as aliases.
3.0.0	28 Aug 2026	New make and model engine covering 73 models across 17 brands, with no API changes; adds fuel type, confidence scores, logo verification and neighbouring countries' plates; served first on the `-v3` endpoints.
2.0	16 Feb 2023	Added body type, year range, orientation, `is_vehicle` and `is_missing_plate`.
1.2	16 Jan 2023	92.28% exact match overall on the 100-image evaluation set.
1.1	23 Dec 2021	86.49% exact match overall on the 118-image evaluation set.
1.0	12 Sep 2021	First release: plate, vehicle type, brand, colour, province.

Request​

Response​

Details​

Request

Response

Details