Changelog - PlateToVIN

It's been a while since PlateToVIN released a new endpoint. We're happy to announce that after months of work, we've released our biggest update yet!

TL;DR

Vehicle Images - We've recreated our images from the ground up, improving quality, dropping response times and improving vehicle matching

VIN Decoding - We've reworked how our VIN decoder makes trim-level decisions. Trim-level matching is more accurate, and as a bonus, we've cut response times in half!

Vehicle Recalls API - We've released a vehicle recalls API that works by VIN or year/make/model, based on the latest data from the NHTSA

License Plate OCR - Over the last few years, we've had countless requests for a license plate OCR. It's still in its early stages so you can expect some hiccups, but we're pretty happy with it

Vehicle Images, rebuilt from the ground up

Every image in our catalog has been regenerated with a new pipeline. The results are sharper, vehicles are matched to your query more reliably, and images are served faster than before.

The basics haven't changed - every image is still a transparent PNG rendered at a fixed, consistent framing, so vehicles line up perfectly across your page. Five angles, six colors, three sizes, all from a single HTTP GET.

If you're already using the image API, there's nothing to change on your end. You should simply notice better images showing up in the same tags.

Smarter VIN decoding, twice as fast

We've rebuilt the way the decoder picks a trim. Instead of leaning on text similarity, it now scores every candidate trim against the vehicle's actual build data and picks the strongest match - which means fewer "close but wrong" trims, especially on model lines with long trim lists.

The request and response formats are identical, so this is a drop-in improvement: same endpoint, same fields, better answers. And with the rework, response times have been cut roughly in half.

New: Safety Recalls API

POST /api/recalls returns open NHTSA recall campaigns for a vehicle. Send a VIN and we'll decode it in-house and return every campaign filed for that year, make and model - newest first - or query by year, make and model directly.

Each campaign includes the affected component, a summary of the defect, the safety consequence, the remedy, and flags worth surfacing to drivers: park_it (do not drive), park_outside (fire risk) and over_the_air_update.

Recall lookups are free per call on plans that include them - they never consume credit and have no monthly cap. One thing to know: recall campaigns apply to a vehicle configuration (year/make/model), not to an individual VIN, so whether a specific vehicle has already been repaired is manufacturer-held data and isn't included.

New: License Plate OCR (early access)

The one you've been asking for. POST /api/ocr takes a photo - a full scene is fine, no cropping needed - finds the plate, reads it, corrects it against the issuing state's plate format, and returns:

The corrected plate, plus a display version with the state's separators
A ranked list of candidate states with probabilities
Whole-string and per-character confidences, plus the plate's bounding box
Plate type and a vanity-plate flag

If you know roughly where a photo was taken, pass state_suggestions (e.g. ["NJ","NY"]) - it meaningfully improves accuracy. Every response includes a trusted flag; when a read is trusted, feed the plate and state straight into /api/convert to resolve the VIN. Blurry or uncertain reads come back clearly marked as low confidence rather than pretending certainty.

A note on privacy: uploaded images are never stored. They're processed and discarded.

OCR is in its early stages - you may hit the occasional misread, and we're actively improving it. Plans include a monthly OCR allowance with per-call overage beyond it.

Full details, parameters and code examples for everything above are in the API documentation. If you hit anything odd - especially with OCR - let us know.