Medical reasoning, with the whole record in view.
Premedice 4 finds the changes that are easy to miss, answers with sources, and keeps uncertainty visible. One model. One endpoint.
Decision support, not a diagnosis. Clinical decisions need a licensed clinician and primary sources.
The change a short visit can miss.
Premedice follows the record between visits and puts the signal in front of the patient and the clinician.
Why Premedice
Premedice 4 is built around the evidence, the uncertainty and the clinical question in front of you.
Follows changes across labs, medicines, symptoms and history, not just a single result.
Every answer arrives with sources attached, so the evidence can be checked.
Labs, medicines, documents, symptoms and history are considered together.
Interactions, pharmacogenomic notes and cautions are surfaced with the record.
Capabilities
Medical reasoning with the record, the sources and the uncertainty in view.
Treatment, contraindications, monitoring and the evidence behind the answer.
Values, ranges and changes across reports, explained in plain language.
Interactions, pharmacogenomic notes and cautions with the record.
A focused summary for a visit, review or second opinion.
Comparison
Exam-style accuracy, not clinical validation.
| System | MedQA | Benchmark | Source |
|---|---|---|---|
| Premedice 4 | 84.6% | Internal MedQA replication | Premedice internal (not peer-reviewed) |
| Med-Gemini | 91.1% | MedQA | Google Research, 2024 |
| Med-PaLM 2 | 86.5% | MedQA | Nature, 2023 |
| MedGemma 27B | 87.7% | MedQA | Google, 2025 |
MedQA is an exam-style benchmark. It is not a clinical validation or an outcome study. Scores are not directly comparable across different test sets. Premedice's 84.6% is an internal replication and has not been peer-reviewed.
Specifications
Safety boundaries
Premedice 4 is decision support. Clinical decisions belong with a licensed professional who knows the patient.
One call
curl https://api.premedice.com/v1/chat/completions \
-H "Authorization: Bearer pm_live_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "Premedice/Premed-4",
"messages": [
{"role": "user", "content": "What changed in the 2025 hypertension guidelines?"}
]
}'The response includes the answer and its sources.
Read the API docsQuestions
Premedice 4 is a proprietary medical reasoning model built for earlier warning tracking, cited answers and whole-record context.
One OpenAI-compatible chat completion. The response includes the answer and its sources.
No. Premedice 4 is decision support, not a diagnosis or treatment device. Clinical decisions need a licensed clinician and primary sources.
Yes. Point the base URL at https://api.premedice.com/v1, use your Premedice key, and set the model to Premedice/Premed-4.
Put it to work