#voice-agents

Mistral's Voice Agents SDK runs transcription, a text model and speech as one managed pipeline at $0.04 or $0.08 a minute. Browser vs phone, tools and costs.

Mistral's Voxtral 3 is four separate speech models in Public Preview. What each does, what it costs per minute or character, and what preview leaves unsettled.

OpenAI now creates a custom voice from a text description for free. Where prompt-made voices work, the 100-voice limit and what it costs to use one.

xAI and OpenAI both upgraded voice AI this week: Grok Voice 3.0 for API voice agents, GPT-Live-1.1 in ChatGPT. Prices, plan limits and the benchmarks.

Test voice agents on mixed-language calls, proper names and language changes. Separate listening errors from reply-language errors and incorrect actions.

Compare voice AI billing without mixing characters, tokens and minutes. Build a useful workload estimate that includes retries, silence and accepted audio.

Tavus's Griffin-Lite joins OpenAI's GPT-Live-1 and three open models that listen while they speak. Latency, price, license and how to test each one yourself.

Google split its live voice model in two on September 15: one answers at once, one reasons while it speaks. Which to pick, and where each is available.

Grok Voice Think Fast 2.0 ships at 0.70s to first audio and $0.08/min on the raw model API. What speech-to-speech buys you, and what a cascade buys you instead.

Qwen-Audio-3.0-TTS-Plus tops the Artificial Analysis TTS arena (~1,236 Elo) at ~$27.59/1M chars, a third of ElevenLabs — but it is hosted-only, no weights.

xAI shipped a no-code Voice Agent Builder on July 1, 2026: a plain-language prompt to a working phone agent in about two minutes, on one $0.05/min meter.

Google's Gemini 3.5 Live Translate streams speech-to-speech in 70+ languages, seconds behind the speaker. A build-vs-buy guide for multilingual CX teams.

Voice agent infrastructure reference — ElevenLabs, Vapi, Retell, and Bland compared on architecture, latency, language support, and real-time turn-taking.