Not just a Toyota. A Land Cruiser 300.
The model classifier distinguishes 2,003 exact models across 145 brands - learning the fine visual differences in grilles, lights and body lines that separate a Camry from a Corolla or an X5 from an X6.
2,003
exact models
145
brands covered
85.2%
model accuracy
Every model we recognize
Search by model name or filter by brand. This is the live class list of the model network - if it’s here, the API can name it.
Showing 30 of 2003 models
| Brand | Model |
|---|---|
| Toyota | 4Runner |
| Toyota | 86 |
| Toyota | AVANZA |
| Toyota | Allex |
| Toyota | Allion |
| Toyota | Alphard |
| Toyota | Altis |
| Toyota | Aqua |
| Toyota | Aurion |
| Toyota | Auris |
| Toyota | Avalon |
| Toyota | Avensis |
| Toyota | Belta |
| Toyota | Blade |
| Toyota | Brevis |
| Toyota | C-HR |
| Toyota | COASTER |
| Toyota | Caldina |
| Toyota | Camry |
| Toyota | Camry Gracia |
| Toyota | Camry IX |
| Toyota | Carina |
| Toyota | Celica |
| Toyota | Chaser |
| Toyota | Corolla |
| Toyota | Corolla Axio |
| Toyota | Corolla Ceres |
| Toyota | Corolla Cross |
| Toyota | Corolla Fielder |
| Toyota | Corolla Rumion |
Two networks, one honest verdict
The final answer is not a single network’s guess. A 181-class brand specialist and the 2,003-class model network each make an independent prediction, and calibrated confidence gates fuse them into one make_model verdict.
When the two networks disagree and neither is confident, the API withholds the answer - brand_uncertain is set to true and the label comes back null. A decision field tells you which fusion rule fired, so uncertain results are always distinguishable from confident ones.
- Agreement between both networks produces the highest-confidence verdicts.
- A confident brand specialist can veto an implausible model prediction.
- Unresolvable conflicts return null - never a silent wrong answer.
The hardest task in vehicle AI
Telling 2,003 model classes apart is far harder than reading a badge: the network must learn generation-level styling cues across millions of photos. It scores 85.2% overall accuracy, measured on hundreds of thousands of real listing photos the network never saw during training.
Frequently asked questions
What format do model predictions come in?
Each prediction is a brand and model pair - for example Toyota with Land Cruiser 300, or BMW with X5 - plus a confidence score and the top-5 candidate list. The fused make_model object also reports which network decided and why.
What happens if my exact model is not covered?
The brand classifier still covers 181 makes, so you get a reliable brand verdict even when the model network has no matching class. You can also request coverage - the class list grows with every retraining cycle.
Can it tell apart similar models like X5 and X6?
Yes - that is exactly what the network is trained for: fine-grained differences in grilles, lights, rooflines and proportions. For close calls, use the confidence score and top-5 candidates to decide how to handle the result.
Does it distinguish model generations?
Some classes are generation-specific (for example Land Cruiser 200 Series vs Land Cruiser 300), and the training data spans multiple generations per model. Where generations share a class, the API returns the common model name.
How do I get only brand and model without the other analysis?
Call /v1/car with body_type=false, color=false and position=false. Only the brand and model networks run, and the response returns faster.
Start recognizing cars today
Free tier included - 100 free credits every month, no credit card required. Your first car profile is one POST request away.