[Breaking News] Google Launches "Gemini 3.8 Live," Going Head-to-Head with OpenAI's Voice AI on Price

News has just broken that makes me feel like AI capable of voice conversation has taken another significant step forward. The new voice AI model "Gemini 3.8 Live," announced by Google on September 15, 2026, is going head-to-head with OpenAI's "GPT-Live-1," which was released just five days earlier on September 10, and a very intense competition has begun in terms of both price and performance. I am learning how to use AI agents and voice AI at Degihaku Generative AI, and with new voice models from the two major companies appearing in such quick succession, I imagine many people are wondering which one they should choose. In this article, I will organize what happened with "Gemini 3.8 Live" and how it differs from "GPT-Live-1" from my perspective.

What happened with Google's "Gemini 3.8 Live"

On September 15, 2026, Google announced "Gemini 3.8 Live," a new voice AI model capable of continuous, natural, real-time conversation, and a higher-end version, "Gemini 3.8 Live Extended Thinking," which supports more complex reasoning. A major feature is its ability to carry on uninterrupted conversations in 97 languages while simultaneously handling tasks like searching and tool invocation in the background. It is said to be designed with a focus on reducing the unnatural pauses that occur while speaking. It is already available for developers via Google AI Studio and the Gemini Live API, and a preview has begun for enterprise users via Gemini Enterprise. It also appears to be already active for end-users in voice search and within the Gemini Live app.

The pricing is also quite aggressive, with voice input set at $0.005 per minute and voice output at $0.018 per minute. Calculating the total for two-way interaction comes to about $0.023 per minute, which is approximately $1.38 per hour. The higher-end Extended Thinking version, used for complex processing, is priced at about $3.50 per hour, offering a two-tier structure depending on the intended use.

Main Specifications

To summarize the key points again, they are as follows:

• Announcement date: September 15, 2026
• Supported languages: 97 languages
• Pricing (Standard version): $0.005 per minute for voice input, $0.018 per minute for voice output
• Higher-end version: Gemini 3.8 Live Extended Thinking (for complex reasoning, approx. $3.50 per hour)
• Availability: Google AI Studio, Gemini Live API, Gemini Enterprise (preview)

How does it compare to OpenAI's "GPT-Live-1"?

Actually, I feel that this "Gemini 3.8 Live" was launched with a timing that strongly considers "GPT-Live-1," which OpenAI announced just five days earlier on September 10. GPT-Live-1 is a full-duplex voice model that allows you to speak while listening, and it had just begun API availability at a price of $0.05 per minute. GPT-Live-1 itself is designed to handle conversation, pacing, and interruptions, while leaving complex reasoning and tool execution to backend models like GPT-6 Astra, and it was released with a choice of 12 different voices. Comparing them as full-duplex voice AIs in terms of price and supported languages makes the differences clear.

| Model | Provider | Announcement Date | Estimated Price |

| GPT-Live-1 | OpenAI | September 10, 2026 | $0.05 per minute |

| Gemini 3.8 Live | Google | September 15, 2026 | Approx. $0.023 per minute |

| Grok Voice Think Fast 2.0 | xAI | 2026 | $0.08 per minute |

Looking at this, you can see that Gemini 3.8 Live was launched at less than half the price of GPT-Live-1. Moreover, Gemini 3.8 Live is said to support 97 languages, and its broad multilingual support, aimed at global expansion, is also a strength. On the other hand, GPT-Live-1 has a reputation for the natural pacing cultivated in ChatGPT's voice features, and its appeal lies in the flexibility to be freely combined with high-performance reasoning models like GPT-6 Astra. I feel that it cannot be said that it is superior simply because it is cheaper; the choice depends on what kind of conversation you want to entrust to it.

The reason why competition in voice AI has become this intense

The reason voice AI is receiving so much attention is that it has become realistic to replace it in practical applications such as call centers, reservation handling, language learning, and medical interview support. The very point of contact with AI is shifting from a way of asking questions in text and reading answers to a way of speaking naturally as if on a phone and receiving an immediate response. That is precisely why OpenAI and Google are trying to take the initiative in both price and performance, and with xAI's Grok Voice included, the major players are all launching new models around the same time. For a while, it seems likely that a price-cutting war and a performance competition will continue in parallel, centered on these three companies.

What readers need to know now

Hearing stories like this might make it seem like a topic only for developers, but in reality, these voice AIs should soon be running behind the scenes of the services we use every day. There is a possibility that customer support phone lines and app voice assistants will become increasingly natural in their responses over the next few months, and for corporate representatives, this price difference cannot be ignored as a factor when comparing implementation costs. As an individual, you don't need to change anything immediately, but if you keep track of which models have what strengths, the speed of your decision-making will change when the time comes to consider implementation. I personally try to touch and compare each new model as it comes out at DigiHaku Generative AI, and I have realized that the actual experience is quite different from just looking at the numbers.

Summary

To me, the fact that Gemini 3.8 Live and GPT-Live-1 were launched just five days apart symbolizes that the voice AI field has truly entered a critical stage. Google has taken a significant lead in terms of price, but when considering the inference models that can be combined with them and the naturalness of the actual conversation, my honest impression is that it is still too early to decide which one is the favorite. I intend to continue following each new model as it is released, including not just the pricing but also the actual user experience.

> For those who want to learn more, at DigiHaku Generative AI, where I am learning while actually using the tools, you can systematically learn about the trends of these latest models and practical ways to use AI. If you are interested, please check it out.

*The information in this article is current as of September 17, 2026.

#GenerativeAI #AINews #Gemini #GPTLive #Google #OpenAI #VoiceAI #AIUtilization #DigiHaku #Technology