Deepgram ships Nova-3 Pharma for drug-name transcription (Sep 17, 2026)

What you get

As of September 17, 2026, Deepgram's nova-3-pharma model gives more accurate drug-name recognition for pharmacy and healthcare voice workflows.

Prerequisites

A hosted Deepgram account and API key in DEEPGRAM_API_KEY (self-hosted access is by request)
English audio (en, en-US, en-AU, en-CA, en-GB, en-IE, en-IN, en-NZ)

Steps

Install the Python SDK:

Shell

pip install deepgram-sdk python-dotenv

Use the standard pre-recorded call and set model to nova-3-pharma (works for streaming too):

Python

import os
import dotenv
from deepgram import DeepgramClient

dotenv.load_dotenv()

deepgram = DeepgramClient(api_key=os.getenv("DEEPGRAM_API_KEY"))
response = deepgram.listen.v1.media.transcribe_url(
    url="https://dpgr.am/bueller.wav",
    model="nova-3-pharma",
    smart_format=True,
)
print(response)

Test it

Point url at a recording that mentions medication names and compare the output with model="nova-3". Drug names should come back spelled correctly more often.