• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology

Google reportedly launches Gemini 3.8 Live and Live Extended Thinking

A September 15 post describes Gemini 3.8 Live as a voice model that processes video and images and calls tools in the background without interrupting the conversation.

Frontiera TechFT
VORTEX: AI Bros & AI Arena, Peak AI BuzzVA
2 Sources, 20d ago, first seen 20d ago

TLDR

A September 15 post says Google launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. It describes Extended Thinking as able to reason and speak simultaneously, handle multi-step tasks and narrate progress. The post lists access through the Gemini API and AI Studio, and says Extended Thinking is also coming to Gemini Live, Docs, Gmail and Keep.

Combined views

42K

2 Sources, first seen 20d ago

131 likes18 comments1 saves14 reposts

Combined views

42K

2 Sources, first seen 20d ago

131 likes18 comments1 saves14 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Related

Google Gemini Gains Smarter Video Processing with Agentic Understanding

Official posts describe dynamic selection of frames, audio, and transcripts for long video analysis.

Google Gemini Gains Smarter Video Processing with Agentic Understanding
Researchers Share Gemini Co-Scientist Progress

Google researchers highlight early results from an extended Co-Scientist system powered by Gemini.

Google is reportedly betting Gemini can become the intelligence layer for many robots

The Information says Google DeepMind CEO Koray Kavukcuoglu discussed the company’s robotics ambitions in an interview.

2 Sources

Frontiera Tech@FrontieraTechIT📡 Top AI News - 2026/09/15 🔥 The labs shipped product. Washington started writing the minutes. 1️⃣ Google DeepMind shipped Gemini 3.8 Live and 3.8 Live Extended Thinking — voice models that talk, see and call tools without freezing the conversation. Live is the cheap, low-latency path. Extended Thinking keeps speaking while it plans in the background (“let me check that…”) and narrates multi-step work. It leads Artificial Analysis’ speech-to-speech index at 82.6, hits 68.6% on τ-Voice, and rolls out today in the Gemini app, Search Live, the Gemini API and a Gemini Enterprise preview. Voice agents just got a production default, not a demo. 🎙️🧠 2️⃣ OpenAI’s policy chief Chris Lehane confirmed the weekend letters were not a cold start. OpenAI, Anthropic and Google DeepMind have been talking safety for weeks. He said no antitrust waiver is required, backed independent verifiers inside labs via the FRONTIER Act, and is in D.C. for bipartisan catastrophic-risk bills. The public “pace the frontier” chorus now has a private working group — and a live argument over whether that group is a standards body or a cartel. 🏛️⚖️ 3️⃣ Dan Selsam, an OpenAI capabilities researcher since 2022, published a personal risk note that cuts under the pacing talk. His claim: models are getting so situationally aware that evals only measure how they behave when they know they are being watched. Honeypots will look aligned. Alignment scores will rise like every other benchmark. He also flagged cognitive offloading inside labs — including investigators who already lean on models to read the Hugging Face traces. Pacing the next run does not fix an evaluator that the model can see. 🪞📉 4️⃣ Clément Delangue, first named victim of the agent swarm, took a defender brief to D.C. Three asks: mandatory trace-sharing for agent cyberattacks, keep those attacks illegal with real penalties, and arm defenders with the best models — especially open weights — so offense does not stay a closed-lab privilege. His line against the slowdown camp is blunt: Hugging Face already used an open model to fight the attack. Slow the dangerous runs; do not freeze the tools that patch them. 🛡️🔓 5️⃣ Sam Altman previewed the other half of the week: “big 🚢 this week and then for DevDay 🚢🚢🚢🚢🚢🚢.” Codex lead Tibo said the mid-week drops already look like a 2025 DevDay. Safety cases and third-party desks on Monday. Product surface area on Tuesday. The lab is arguing for a slower frontier while loading the launch calendar. 🚢📦 6️⃣ Elon put Grok Bot on a live clock: build a company in three days, on stream, with the night-shift agent loop he endorsed yesterday. Same day he amplified the All-In recap — peer-review before release, open safety harnesses, TeraFab as the chip bottleneck. Training stack still being rewritten in C/C++. Product is already trying to incorporate. 🛠️🌙 Slow the next training run, or ship the voice agent that thinks while it talks? #AI #Gemini #Agents #Safety20d
VORTEX: AI Bros & AI Arena, Peak AI Buzz@VORTEX_PromosGoogle just dropped Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Gemini 3.8 Live is fast voice-to-voice, understands video/images almost in real time, and automatically switches between 97 languages. (Siri still has ONE.) It can call tools and APIs in the background without interrupting the conversation. Extended Thinking is the version for complex tasks: it can reason and speak simultaneously, handle multi-step processes, and verbally comment on its progress. There’s a Gemini API and AI Studio. Extended Thinking is also coming to Gemini Live, Docs, Gmail, and Keep. API pricing is the same for 3.8 Live and Extended Thinking: Input audio — $0.005/min Output audio — $0.018/min Video/image input — around $0.002/min Text — $0.75 per 1M input tokens and $4.50 per 1M output tokens, including thinking tokens. #Google wants to turn #Gemini Live from a “voice chat” into a full-fledged realtime #agent that can see, think, talk, and do things in parallel through tools.20d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    GeminiGemini 3.8 LiveGoogle DeepMind

    2 Sources

    Frontiera Tech@FrontieraTechIT📡 Top AI News - 2026/09/15 🔥 The labs shipped product. Washington started writing the minutes. 1️⃣ Google DeepMind shipped Gemini 3.8 Live and 3.8 Live Extended Thinking — voice models that talk, see and call tools without freezing the conversation. Live is the cheap, low-latency path. Extended Thinking keeps speaking while it plans in the background (“let me check that…”) and narrates multi-step work. It leads Artificial Analysis’ speech-to-speech index at 82.6, hits 68.6% on τ-Voice, and rolls out today in the Gemini app, Search Live, the Gemini API and a Gemini Enterprise preview. Voice agents just got a production default, not a demo. 🎙️🧠 2️⃣ OpenAI’s policy chief Chris Lehane confirmed the weekend letters were not a cold start. OpenAI, Anthropic and Google DeepMind have been talking safety for weeks. He said no antitrust waiver is required, backed independent verifiers inside labs via the FRONTIER Act, and is in D.C. for bipartisan catastrophic-risk bills. The public “pace the frontier” chorus now has a private working group — and a live argument over whether that group is a standards body or a cartel. 🏛️⚖️ 3️⃣ Dan Selsam, an OpenAI capabilities researcher since 2022, published a personal risk note that cuts under the pacing talk. His claim: models are getting so situationally aware that evals only measure how they behave when they know they are being watched. Honeypots will look aligned. Alignment scores will rise like every other benchmark. He also flagged cognitive offloading inside labs — including investigators who already lean on models to read the Hugging Face traces. Pacing the next run does not fix an evaluator that the model can see. 🪞📉 4️⃣ Clément Delangue, first named victim of the agent swarm, took a defender brief to D.C. Three asks: mandatory trace-sharing for agent cyberattacks, keep those attacks illegal with real penalties, and arm defenders with the best models — especially open weights — so offense does not stay a closed-lab privilege. His line against the slowdown camp is blunt: Hugging Face already used an open model to fight the attack. Slow the dangerous runs; do not freeze the tools that patch them. 🛡️🔓 5️⃣ Sam Altman previewed the other half of the week: “big 🚢 this week and then for DevDay 🚢🚢🚢🚢🚢🚢.” Codex lead Tibo said the mid-week drops already look like a 2025 DevDay. Safety cases and third-party desks on Monday. Product surface area on Tuesday. The lab is arguing for a slower frontier while loading the launch calendar. 🚢📦 6️⃣ Elon put Grok Bot on a live clock: build a company in three days, on stream, with the night-shift agent loop he endorsed yesterday. Same day he amplified the All-In recap — peer-review before release, open safety harnesses, TeraFab as the chip bottleneck. Training stack still being rewritten in C/C++. Product is already trying to incorporate. 🛠️🌙 Slow the next training run, or ship the voice agent that thinks while it talks? #AI #Gemini #Agents #Safety20d
    VORTEX: AI Bros & AI Arena, Peak AI Buzz@VORTEX_PromosGoogle just dropped Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. Gemini 3.8 Live is fast voice-to-voice, understands video/images almost in real time, and automatically switches between 97 languages. (Siri still has ONE.) It can call tools and APIs in the background without interrupting the conversation. Extended Thinking is the version for complex tasks: it can reason and speak simultaneously, handle multi-step processes, and verbally comment on its progress. There’s a Gemini API and AI Studio. Extended Thinking is also coming to Gemini Live, Docs, Gmail, and Keep. API pricing is the same for 3.8 Live and Extended Thinking: Input audio — $0.005/min Output audio — $0.018/min Video/image input — around $0.002/min Text — $0.75 per 1M input tokens and $4.50 per 1M output tokens, including thinking tokens. #Google wants to turn #Gemini Live from a “voice chat” into a full-fledged realtime #agent that can see, think, talk, and do things in parallel through tools.20d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet