# Google Gemini 3.8 Live Models Top Speech Benchmarks

**Published:** 2026-09-17T13:27:16.926Z  
**Topic:** Google  
**Sentiment:** neutral  
**Publisher:** TrendWatcher — https://www.trendwatcher.in/article/e0266600-1fbe-4960-ab30-8528c546375a

Google's new Gemini 3.8 Live and 3.8 Live Extended Thinking models lead speech benchmarks, with Extended Thinking scoring 82.6 on Artificial Analysis's index.

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, positioning them as its most advanced audio models capable of talking and reasoning [1]. The Extended Thinking variant achieved an 82.6 score on Artificial Analysis’s Speech-to-Speech Index, surpassing competitors like GPT-Live-1 and Grok Voice [1].

| At a glance        |                                       |
| :----------------- | :------------------------------------ |
| Gemini 3.8 Live ET | Speech-to-Speech Index: 82.6          |
| Gemini 3.8 Live    | Speech-to-Speech Index: 76.0          |
| Gemini 3.8 Flash   | Pricing: $0.75/1M input tokens        |
| Gemini 3.8 Flash   | Pricing: $3.75/1M output tokens       |

## Performance Benchmarks and User Preference

Artificial Analysis’s index evaluates speech reasoning, agentic performance, arena preference, and task success rate [1]. Gemini 3.8 Live Extended Thinking debuted at the top with a high reasoning effort, followed by GPT-Live-1 Astra (81.5) and Grok Voice Think Fast 2.0 High (81.3) [1]. The standard Gemini 3.8 Live model secured fifth place with a 76.0 score, both outperforming the previous Gemini 3.1 Flash Live High (71.5) [1].

On specific benchmarks, Extended Thinking reached 68.6% on the Tau Voice benchmark, a significant increase from the previous generation's 37.7% [1]. Its speech reasoning score was 97.7%, narrowly trailing Qwen Audio 3.0 Realtime Plus at 99.2% [1].

However, user preference metrics show a different trend. In blind Speech Agent Arena conversations, Gemini 3.1 Flash Live still leads with an Elo of 1096, while Gemini 3.8 Live is second at 1083, and Extended Thinking trails at 990 despite completing 89.1% of its tasks [1]. This suggests a divergence between models that excel in complex reasoning and those preferred for conversational flow [1].

## Pricing and Accessibility

The standard Gemini 3.8 Live model offers competitive pricing at $0.84 per hour of input audio, approximately half the cost of its predecessor [1]. The Extended Thinking variant is priced at $3.50 per hour, undercutting GPT-Live-1 Sol ($4.47) and Grok Voice Think Fast 2.0 High ($4.80) [1]. Latency has also improved, with average time to first audio dropping to 1.18 seconds from 2.99 seconds for the older Gemini model [1].

Gemini 3.8 Flash, released separately, maintains the same introductory pricing as its predecessor: $0.75 per million input tokens and $3.75 per million output tokens [3]. Google warns that this model may use more tokens for optimal performance, potentially increasing overall costs for users [3]. This version is available to Google AI Pro and Ultra subscribers and developers [2, 3]. A specialized variant, Gemini 3.8 Flash Cyber, is also launching to Google’s Fairwind Program for governments and trusted partners, focusing on cybersecurity capabilities [3].

## What to watch

*   Google’s rollout of Gemini 3.8 Flash to the free tier of AI Mode.
*   The performance and pricing of upcoming models like Gemini 4.
*   Adoption rates of Gemini 3.8 Live Extended Thinking in applications requiring high reasoning.

The dual focus on benchmark performance and user preference highlights a developing dynamic in voice AI, with Google leading in both areas through different models [1]. The accessibility of Gemini 3.8 Flash, with its tiered performance modes and competitive pricing, positions it for broad adoption in practical applications [4].

## Sources
1. BeInCrypto — [Google Just Released Its Most Advanced Audio Model. Here Is How It Ranks](https://beincrypto.com/gemini-3-8-live-speech-benchmark/)
2. Searchenginejournal.com — [Google Adds Gemini 3.8 Flash To AI Mode](https://www.searchenginejournal.com/google-gemini-3-8-flash-ai-mode/588194/)
3. The Verge — [Google says its new Gemini 3.8 Flash model ‘works harder’ but might cost more](https://www.theverge.com/ai-artificial-intelligence/988742/google-gemini-3-8-flash)
4. Geeky Gadgets — [Gemini 3.8 Flash Matches Opus 5 on Deep Sweep 1.1 Benchmark](https://www.geeky-gadgets.com/google-gemini-3-8-flash/)

---
Cite as: TrendWatcher, "Google Gemini 3.8 Live Models Top Speech Benchmarks", https://www.trendwatcher.in/article/e0266600-1fbe-4960-ab30-8528c546375a