AI CRITIQUE

Görsel, mavi degrade arka plan ve stilize bir blok grafik eşliğinde Gemini 3.8 Live ile 3.8 Live Extended Thinking özelliklerini tanıtıyor.

Google releases Gemini 3.8 Live models capable of real-time coding

Read the article English (automatic translation)

Summary

Google has released Gemini 3.8 Live (cost-effective) and Gemini 3.8 Live Extended Thinking (advanced multi-step reasoning), real-time voice chat models. Both models support 97 languages and offer live coding capabilities during verbal dialogues. Third-party benchmarks report that the Extended Thinking variant outperforms OpenAI's GPT-Live-1-Astra and SpaceXAI's Grok Voice Think Fast 2.0 models. The base model placed second in the blind human evaluation arena.

Why it matters

Real-time voice coding bridges the gap between verbal interaction and generative software development. By allowing developers and non-technical users to iteratively refine code without switching to a text-based IDE, it could reshape dynamics in this field. Support for 97 languages and the ability to invoke tools in the background without interrupting speech are seen as moves toward multilingual, hands-free AI assistants that not only converse but also act. In the broader AI race, a reliable third-party benchmark victory where a model from OpenAI and SpaceX brands is outperformed by both intensifies competition in the voice agent segment.

Highlights

Both models support 97 languages and offer background tool invocation without interrupting ongoing voice conversations.
In the Speech Agent Arena blind test, another Google model (Gemini 3.1 Flash Live Minimal) took first place, while 3.8 Live came in second.
The Extended Thinking variant is positioned for multi-step reasoning and was demonstrated as a real-time programming tutor.

Critical take

The 'second place' title in the Speech Agent Arena feels somewhat weak given that the first-place finisher is also a Google model. This implies the arena effectively crowned the Gemini family twice. Additionally, the 'real-time coding' capability is only demonstrated in a selected promotional video. The article does not provide independent, repeatable benchmarks regarding live coding accuracy or latency under load, leaving the practical reliability of this feature largely unverified.

Bu kritik, yerel yapay zeka modeli Qwen3.8-27B tarafından yazılmıştır. Haberi ben seçtim; kritik metnini yayımlanmadan önce tek tek denetlemiyorum.This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.

Back to the issue