STAYCHARTED AMT / PRODUCT
Compare AMT’s text and picture models: what they learn, when to try them, and which plans include them.
START WITH YOUR DATA
Learn your categories from labeled examples. Compare the errors before choosing a model.
Plan availability below describes which models you can train. Check Pricing for current limits and Vision add-on details.
| Model | Reads | How it learns | Available on |
|---|---|---|---|
| Classifier Model | Text | Words and phrases from your examples | Every plan |
| AI Classifier Model | Text | Words plus meaning from a pre-trained language model | Free (1), Essentials and Business |
| Dedicated AI Model | Text | Fine-tunes a language model on your examples | Business |
| AI Image Classifier Model | Pictures | Pre-trained visual representations plus your categories | Essentials or Business with Vision |
| Dedicated AI Image Model | Pictures | Fine-tunes a vision model on your pictures | Business with Vision |
TEXT
Learns which words and phrases go with your categories. Start here for text with repeated terminology and clearly separated labels. It can miss a familiar meaning expressed with unfamiliar words. Available on every plan.
Combines words with representations from a pre-trained language model. It can connect different wording with similar meaning while learning your own categories from labeled examples. The shared pre-trained language model is not fine-tuned on your uploads. Free includes one AI Classifier Model within its three-model allowance.
Fine-tunes an AI language model on your examples. Try it when distinctions depend on context or subtle wording. Training is more involved than an AI Classifier Model, so compare measured accuracy and the errors that remain rather than assuming the larger model will always be better.
PICTURES / VISION
Uses a pre-trained vision model to read pictures and learns your categories from labeled examples. Available with Vision on Essentials and Business. Train from photos in category folders or a spreadsheet of picture links. Vision classifies what pictures show; it is not OCR for extracting receipt or document text.
Fine-tunes a vision model on your labeled pictures. Compare it with the AI Image Classifier Model when your categories depend on details the pre-trained representation does not separate well. Use representative held-out pictures to decide whether the change helps.
Start with an available model and representative examples. Check whether its mistakes are acceptable for the task.
A higher overall score can hide a weak category. Look at the labels your team most needs to get right.
Confidence helps you prioritize checks. It does not guarantee correctness or replace review for consequential uses.
Public benchmarks show how approaches differ on specific datasets. Your own data and labeling rules determine the best fit.
Start with Free for text classification. Compare plans for AI models, Vision, and API access.