Google’s new speech model Gemini 3.8 Live supports real-time reasoning

🤖 AI/ML Daily Signal — Evening Edition 15 Sep 2026 · 17:12 UTC ──────────────────────────────── 🔥 1. Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking 🏭 Google DeepMind Blog · Score: 9/10 Google DeepMind releases Gemini 3.8 Live and 3.8 Live Extended Thinking, enabling real-time streaming interactions and enhanced reasoning capabilities. This directly impacts how practitioners build and deploy conversational AI systems at scale. Read more → ⭐ 2. Task-Aware Federated Fine-Tuning for MoE-based Large Language Models 🔬 arXiv cs.LG (Machine Learning) · Score: 8/10 Proposes task-aware federated fine-tuning methods for MoE-based LLMs, addressing computational efficiency in distributed settings. Directly applicable to organizations training large models across multiple devices while maintaining performance. Read more → ⭐ 3. AttnFuse: A Composable DSL for Compiling Attentions to Fused GPU Kernels 🔬 arXiv cs.LG (Machine Learning) · Score: 8/10 AttnFuse introduces a DSL for compiling attention mechanisms to fused GPU kernels, reducing computation and memory overhead. Critical infrastructure work that enables faster, more efficient transformer inference in production. Read more → ✅ 4.

Ⓜ️ Google’s new voice models just topped the speech-to-speech leaderboard Google has released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, targeting real-time voice agents rather than traditional text-based interactions. Gemini 3.8 Live reportedly reaches frontier-level performance at around $0.84 per hour, while delivering a higher composite score than GPT-Realtime-2 High at roughly 80% lower measured cost. Both models are multimodal: voice is the main interface, but they can also process visual information during a conversation and automatically switch between 97 supported languages. The biggest difference comes with Extended Thinking. It scores 68.6% on τ-Voice, narrowly ahead of GPT-Live-1 Astra Medium at 67.9%. τ-Voice isn’t simply testing whether an AI sounds natural. It measures whether a voice agent can actually complete complex, multi-step customer-service tasks while following policies, using tools correctly and reaching the right outcome across airline, retail and telecom scenarios. @aipost 🏴

🔍 Google adds background reasoning to Gemini Live Gemini 3.8 Live can keep a voice conversation going while it calls tools and handles multi-step tasks through APIs. It also processes images almost in real time and switches between 97 languages during a call. The Extended Thinking version targets harder voice tasks. It ranked first on Speech to Speech Quality Index with 82.6 and scored 97.7% on Big Bench Audio. Both models are appearing in Gemini API and Google AI Studio. Extended Thinking is also available in Gemini Live and some Google apps. 📊@tech