# Baseten builds fast dental note‑taking model matching Gemini 2.5‑pro

> Baseten fine‑tuned a low‑latency LLM for dental documentation that handles long transcripts, real‑time updates, and tooth notation, running 5‑6× faster than Gemini 2.5‑pro.

Oossa · 2026-10-03 · https://oossa.com/en/baseten-builds-fast-dental-note-taking-model-matching-gemini-2-5-pro

Baseten worked with a leading dental note‑taking company to create a purpose‑built language model that writes structured dental notes from up to 25,000‑token transcripts. The model learned three tasks – full note generation, real‑time addition of findings, and note alteration – using a single LoRA adapter. In tests it matched Gemini 2.5‑pro’s accuracy but was 5‑6 times faster, and achieved 90% accuracy on tooth‑notation checks.

## The facts

- Model handles transcripts up to 25,000 tokens
- Latency 5‑6× lower than Gemini 2.5‑pro while matching its accuracy

## Why it matters

Dentists can get accurate, real‑time documentation without the lag of larger, slower models, freeing time for patient care.

## Sources & references

1. [Purpose-built LLMs for dental note-taking](https://www.baseten.co/blog/purpose-built-llms-for-dental-note-taking/) – Baseten

Last updated: 2026-10-03