GPT-Live Fixes ChatGPT's Voice Cadence. xAI Already Priced the Infrastructure Version.

OpenAI replaced the default ChatGPT Voice experience on July 8 with GPT-Live, built on a full-duplex architecture that lets the model listen and speak at the same time instead of taking turns — it can backchannel with “mhmm” or “yeah,” handle interruptions mid-sentence, and go quiet while a user is still thinking. When a request needs real reasoning or web search, GPT-Live hands off to GPT-5.5 behind the scenes and folds the answer back into the live conversation once it’s ready. Two tiers ship at once: GPT-Live-1 as the default for paid ChatGPT tiers, GPT-Live-1 mini for the free tier worldwide.

The turn-taking problem GPT-Live is solving — the pause-and-wait cadence that makes AI voice feel like a phone tree — is the same one xAI priced a solution for a week earlier. Grok’s Voice Agent Builder, in beta since July 1, lets a company configure a production voice agent from a plain-language prompt in about two minutes, bundling telephony, a document-based knowledge base, and CRM tool-calling, at a flat $0.05 per minute with no platform fee. xAI’s own benchmark has Grok Voice scoring 67.3% against Gemini 3.1 Flash Live’s 43.8% and GPT Realtime 1.5’s 35.3% — a number worth the usual skepticism reserved for a vendor grading its own homework, but directionally it says the industry has converged on “does this feel like a real conversation” as the metric that matters.

Two different products are chasing the same outcome from opposite ends: OpenAI is fixing the flagship consumer assistant’s cadence, xAI is packaging the same capability as infrastructure a company can wire into its own phone line. For a business deciding where AI voice actually earns its place — customer service, scheduling, lead qualification — the interesting fact isn’t which chatbot sounds more human. It’s that the cost of standing up a passable voice agent dropped to a few cents a minute in the space of a single week.