Hi VAPI team,

I'd love to see native support for Mistral's Voxtral model family as a provider option across all three pipeline layers:

• STT: Voxtral Realtime (sub-200ms latency, 13 languages, open-weights Apache 2.0)

• LLM: Voxtral Small 24B (function-calling directly from voice, multilingual)

• TTS: Voxtral TTS (4B params, very low latency, 9 languages, zero-shot voice cloning)

What makes Voxtral particularly interesting for VAPI users is that it's the first open-source stack that covers all three layers under a single model family

Voxtral Realtime already achieves near-offline accuracy at 480ms delay, which fits right within VAPI's <700ms voice-to-voice target.

The Apache 2.0 license also makes it attractive for enterprise and regulated industry use cases.

Would love to see Mistral API keys supported at minimum, and ideally a self-hosted endpoint option for Voxtral Realtime via vLLM.

Thanks very much guys !