Last updated October 2026 · 16 documents, every field scored
How Pawssier compares on veterinary records
We ran 16 sample veterinary documents through Pawssier, Claude, ChatGPT and Gemini, then ran them all again. Every field was scored against a hand-checked answer key, and we checked whether each tool gave the same answer the second time. Pawssier matches the best general AI models on accuracy, with no wrong or made-up values, and it gave the same answer the second time more often than any of them.
What Pawssier adds
The best model for each document
Pawssier isn't tied to one AI model. We test several models on every kind of veterinary record and send each document to the one that reads that type best, then apply veterinary rules on top so the results come out the same way every time. As new models come out, we test them the same way.
The same answer every time
Send the same document twice and Pawssier returns the same values 98% of the time. The general AI models we tested matched between 80% and 96% of their values.
One exact format
Dates in one format, weights with one number and one unit, microchips without spaces. Ready for your system with nothing to clean up.
No calculated guesses
Pawssier never works out a date from "1 year" or "in 10 days". If the date isn’t printed, the field stays empty.
How to read this: Each row is one tool. Percentages are out of 100, so 98% means 98 of every 100 values. For wrong and made-up values, zero is best.
Best result, including tiesBars show 0% to 100%.
| Tool | Field accuracyFields read correctlyHigher is better | Wrong valuesConfident answers that don't match the documentLower is better | Made-up valuesAnswers where the document shows nothingLower is better | Same answer twiceValues that came back identical on a second runHigher is better | Exact formatDates, weights and microchips in one clean shapeHigher is better |
|---|---|---|---|---|---|
| Pawssier | 99.4% | 0 | 0 | 98% | 99% |
| Claude (claude-opus-5-5) | 99.2% | 0 | 0 | 80% | 99% |
| ChatGPT (gpt-6-astra) | 99.1% | 1 | 0 | 96% | 94% |
| Gemini (gemini-4-argon) | 98.4% | 1 | 2 | 89% | 96% |
Pawssier
- Field accuracyFields read correctlyHigher is better
- 99.4%
- Wrong valuesConfident answers that don't match the documentLower is better
- 0
- Made-up valuesAnswers where the document shows nothingLower is better
- 0
- Same answer twiceValues that came back identical on a second runHigher is better
- 98%
- Exact formatDates, weights and microchips in one clean shapeHigher is better
- 99%
Claude (claude-opus-5-5)
- Field accuracyFields read correctlyHigher is better
- 99.2%
- Wrong valuesConfident answers that don't match the documentLower is better
- 0
- Made-up valuesAnswers where the document shows nothingLower is better
- 0
- Same answer twiceValues that came back identical on a second runHigher is better
- 80%
- Exact formatDates, weights and microchips in one clean shapeHigher is better
- 99%
ChatGPT (gpt-6-astra)
- Field accuracyFields read correctlyHigher is better
- 99.1%
- Wrong valuesConfident answers that don't match the documentLower is better
- 1
- Made-up valuesAnswers where the document shows nothingLower is better
- 0
- Same answer twiceValues that came back identical on a second runHigher is better
- 96%
- Exact formatDates, weights and microchips in one clean shapeHigher is better
- 94%
Gemini (gemini-4-argon)
- Field accuracyFields read correctlyHigher is better
- 98.4%
- Wrong valuesConfident answers that don't match the documentLower is better
- 1
- Made-up valuesAnswers where the document shows nothingLower is better
- 2
- Same answer twiceValues that came back identical on a second runHigher is better
- 89%
- Exact formatDates, weights and microchips in one clean shapeHigher is better
- 96%
How we tested
- The documents are sample records made for testing, not real patient records.
- Each answer key was written and checked by hand.
- A correct field counts as 1 and a partly correct field as half, pooled across every field in every document.
- Each AI model got the same instructions through its API.
- Wrong values are confident values that don’t match the document; made-up values are values given where the document shows none.
- Same answer twice: every document was sent to each tool twice, and we counted the values each tool returned exactly the same way both times.
- Exact format: every date, weight and microchip was checked against one exact shape, whether or not the value was right.
See it on your own records
Try the demo on a real record, or talk to us about what you’re reading.
Want to see what Pawssier finds in your records?
Records arrive from hundreds of different systems, and none of them can read each other’s. Pawssier reads them all and shows you what’s worth another look.