Speech-to-text Community model

Voxtral-Mini-4B-Realtime-2602

Voxtral-Mini-4B-Realtime-2602, Mistral AI's realtime speech-to-text model (a 970M-parameter causal audio encoder feeding a 3.4B Ministral text decoder; 13 languages; one text token per 80 ms of audio), served on four Blackhole chips (2 x…

Community model. It is published as a tt-model bundle and is not in tt-inference-server's release spec. Tenstorrent has not reviewed the model or its bundle. Serve at your own risk.

tt-model bundles

Bundle	Hardware	Served with	Context	Pushed
ndaly/Voxtral-Mini-4B-Realtime-2602	BH QuietBox 2	vllm-plugin	128K	2026-10-21

From tt-model search --catalog and each bundle's manifest (tt-model info). A bundle lists every hardware profile its manifest defines. Each bundle is published by its owner, and Tenstorrent has not reviewed it. Serve at your own risk.

Details

Kind	Speech-to-text (from the task tag on the bundle's card)
Family	Voxtral (from the start of the name)
License (bundle cards)	apache-2.0
Listed by	tt-model bundle

Install and run

Download the weights and the bundle

tt model pull ndaly/Voxtral-Mini-4B-Realtime-2602

Not reviewed by Tenstorrent. Serve at your own risk.

Serve it on this machine

tt serve ndaly/Voxtral-Mini-4B-Realtime-2602

A tt-model bundle, so tt serve runs it through tt-model. If a bundle needs hardware you do not have, tt says so.

Connect a client with tt launch

Serve the model first, then run one of these. Only clients that work with this model are listed.

Web service. Any chat model

tt launch openwebui --model mistralai/Voxtral-Mini-4B-Realtime-2602

Web service. Any chat model

tt launch anythingllm --model mistralai/Voxtral-Mini-4B-Realtime-2602

Terminal clients need a tool-call parser. Open WebUI and AnythingLLM take any chat model. The rule is tt-cli's (tt launch). This model is not in tt's catalog, so tt warns instead of checking. The listing comes from the bundle's manifest, which says it is served with vLLM.