An AI assistant for business is software that listens, reasons, calls external tools and talks back in natural speech to finish a concrete task - book an appointment, place an order, retrieve a document. Mistral's Voxtral 3 is a fresh example on the voice side: one model family covers transcription, speech synthesis and tool calls through a single streaming SDK.
Mistral just released Voxtral 3 as three parts: Voxtral Transcribe 3 (50+ languages, diarization, end-of-turn detection, Mini variant at 4B parameters for $0.0035/min), Voxtral Mini TTS 3 with emotion-aware voices across 35+ languages, and a voice agent SDK that unifies listening, reasoning, tool calling and speaking in real time. Everything runs on Mistral's EU-hosted API or self-deployed on your own infrastructure, so audio can stay inside your perimeter.
What this changes for a small business
Phone-call automation stops being a Twilio-plus-Whisper-plus-GPT patchwork glued together with timeouts. A clinic receptionist that reads a French referral letter and books a slot (Mistral's own demo), a multilingual service desk, a warehouse voice interface - all collapse into one pipeline you can wire into n8n or a custom agent, close in spirit to the voice-controlled Telegram assistant I built. At $0.0035/min for Mini transcription, a steady 40 hours of calls per month is around $8 of ASR cost - the budget now goes into tool integration, prompts and fallback logic, not the audio model.
Why it matters: Open, EU-hosted voice stacks with native tool calling put live phone agents within reach of shops that could not touch a contact-center platform six months ago. If your line rings after hours in two languages, this is the quarter to prototype.
Source: Hacker News