Med-Gemini
Gemini-based medical models hit 91.1% on MedQA (USMLE) with uncertainty-guided web search; paper 'Capabilities of Gemini Models in Medicine'.
Fine-tunes Gemini with web-search integration and custom encoders. Beats GPT-4V by 44.5% on average relative margin across seven multimodal medical benchmarks and handles needle-in-haystack retrieval over long health records.
- Date
- Monday, 29 April 2024
- Lab
- Kind
- paper
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| MedQA (USMLE) accuracy | 91.1% uncertainty-guided search strategy | company |
arXiv v1 2024-04-29. Authors caution that real-world validation is still needed.
Sources
This record was checked against its sources on 6 October 2026. How we check