# Google's Gemini 3.8 Live Tops an Independent Speech Leaderboard at a Tenth of OpenAI's Voice Price

The Extended Thinking variant scored 82.6 on Artificial Analysis's speech-to-speech index while Google priced audio output at $0.018 per minute against $0.05 for OpenAI's GPT-Live-1.

- Published: 2026-09-17T06:17:58.659Z
- Canonical: https://polylog.news/ai/2026-09-17/google-s-gemini-3-8-live-tops-an-independent-speech-leaderbo
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Polylog editors](https://polylog.news), [The Decoder](https://the-decoder.com/google-launches-gemini-3-8-live-to-take-on-openais-gpt-live-1-at-a-fraction-of-the-cost/)

Google DeepMind released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking through the Gemini application programming interface (API) and AI Studio, aiming at real-time voice agents rather than text chat. The Extended Thinking variant [took first place on the Artificial Analysis Speech-to-Speech Quality Index with 82.6](https://t.me/aipost/8166), ahead of OpenAI's GPT-Live-1 models.

The ranking comes from a third party rather than Google, which makes it more useful than a vendor chart, though Artificial Analysis runs a fixed suite and one index does not establish turn-taking quality, interruption handling or latency under load. Google also reported 68.6 percent on the agentic voice benchmark τ-Voice and 35.1 percent on Sierra's banking variant, figures that have not been independently reproduced. The model supports more than 97 languages and can place API calls in the background while continuing to speak.

The price difference matters more. Google charges [$0.005 per minute of audio input and $0.018 per minute of output](https://the-decoder.com/google-launches-gemini-3-8-live-to-take-on-openais-gpt-live-1-at-a-fraction-of-the-cost/), against $0.05 per minute for GPT-Live-1.

OpenAI made a comparable cut elsewhere in its product line. It [reduced the cost of voice control for Work and Codex agents in the ChatGPT desktop client by close to 60 percent](https://t.me/ai_machinelearning_big_data/10940), giving users roughly 2.4 times more voice time for the same credits, and it now bills the voice stream separately from the computing a task consumes. That separation lets the company change the price of conversation without changing the price of the work itself.

## What this means

Voice has been the last interface where per-minute pricing stayed high enough to block always-on deployment in call centers, field service and support. A roughly tenfold gap between Google and OpenAI on published per-minute rates forces a response, and the usual response is a price cut rather than a capability claim. Contact-center software vendors that resell voice minutes with a markup lose pricing room first, because their customers can read both API price pages. The separation of voice billing from task compute is the structural change worth tracking, since it lets vendors advertise a cheap conversational layer while keeping margin in the agent that does the work.

## What to watch

- Whether OpenAI cuts GPT-Live-1 per-minute pricing, which would confirm that voice has entered the same price competition as text tokens.
- Whether independent testers reproduce the τ-Voice agentic scores, since a leaderboard position on audio quality does not prove an agent completes banking or support tasks.
- Whether enterprises move production voice traffic between providers, the practical test of whether these models are interchangeable at the API level.