Deepgram Flux Multilingual Now Available for Telnyx Voice AI Assistants

29, Apr 2026

Deepgram Flux, the STT model optimized for conversational AI, now supports 10 languages on Telnyx Voice AI Assistants. English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch are all available, with three language modes to match different deployment needs. This shipped the same day as Deepgram's own GA announcement.

What's new

10 monolingual options: Flux now supports English, Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian, and Dutch as individual language selections, each with monolingual-grade accuracy and latency.
Multi language mode: Supports any of the above languages in a single session, with some cost to latency and accuracy compared to a single-language selection.
Auto language mode: Telnyx automatically applies the correct language hint for Deepgram Multi, combining the flexibility of Multi mode with the latency and accuracy of a monolingual configuration.
Same-day GA support: Telnyx shipped multilingual Flux the same day Deepgram launched it. No waiting for a platform integration cycle.
On-network processing: Flux runs inside Telnyx-hosted Deepgram models, keeping audio and inference on the same private backbone.

Why it matters

Voice AI agents serving global audiences had to choose between Flux's low-latency turn detection (English only) and Nova-3's multilingual support. Now Flux covers both, so teams don't trade latency for language coverage.
Auto mode removes the need to configure language per call or per number. Telnyx detects the spoken language and applies the right hint, which means one assistant configuration works across all supported languages.
Same-day GA availability means no integration gap. When Deepgram ships a model, Telnyx makes it production-ready immediately.

Example use cases

Customer support teams running a single assistant for English and Spanish callers, using Auto mode to handle both without manual routing.
European enterprises deploying agents in French, German, Italian, and Dutch from one configuration, with per-language accuracy close to monolingual performance.
Global platforms that previously needed separate English Flux and multilingual Nova-3 pipelines consolidating onto a single model.

Getting started

In Mission Control, navigate to AI then Assistants, and select your assistant.
Open the Voice tab and scroll down to Transcription.
Select Flux as your STT model.
Choose your language mode: a single language, Multi, or Auto.
Save your assistant and test with a live call.


import telnyx

assistant = telnyx.Assistant.create(
    name="Multilingual Support Agent",
    model="moonshotai/Kimi-K2.6",
    instructions="You are a helpful support agent.",
    transcription={
        "model": "deepgram/flux",
        "language": "auto"
    },
)

Learn more in the Transcription Settings docs or the Voice AI Agents product page.