- AI Fire
- Posts
- 🏆 Google Takes #1 Voice AI Again!
In partnership with
Google just rolled out Gemini 3.8 Live Extended Thinking. Interestingly, its High mode also grabbed the #1 spot on major voice benchmarks, edging past GPT Live 1.
What's on FIRE 🔥
IN PARTNERSHIP WITH VULTR
Tables, charts, and diagrams can get lost when PDFs are reduced to plain text for RAG. Vultr's new open-weight models preserve that visual context by indexing document pages as images.
This is reflected in the ViDoRe V3 benchmark, with Prime at #2, Core at #4, and Flash leading the sub-1B category.
AI INSIGHTS
Google just launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking across the Gemini API, Google AI Studio, Gemini app, Google Workspace, and Search. Main differences:
Gemini 3.8 Live focuses on fast, low-latency voice conversations, visual grounding, and cost-efficient voice agents.
Gemini 3.8 Live Extended Thinking adds deeper multi-step reasoning that can continue in the background while the model keeps talking to the user.
Extended Thinking supports asynchronous tool calls for workflows such as travel search, technical support, and complex agents.
Both accept text, images, audio, and video, with a 131K-token input context window and up to 65K output tokens.
Extended Thinking High is #1 overall on Artificial Analysis’ Speech-to-Speech Quality Index, yet not #1 on most benchmarks. Google also reports standard Gemini 3.8 Live reached #2 on Speech Agent Arena.
One interesting feature is that Extended Thinking can keep the conversation alive while it works. For example, it can say it is checking flights, launch several API calls in parallel, reason over the results, and continue speaking when the results arrive.
PRESENTED BY HUBSPOT
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:
7 high-impact AI strategies to accelerate your marketing performance
Practical use cases for content creation, lead gen, and personalization
Expert insights into how top marketers are using AI today
A framework to evaluate and implement AI tools efficiently
Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.
AI SOURCES FROM AI FIRE
1. Free Guide: GPT-6 Astra Gets 10X More Productivity When You Give It This One Type of Data. In about 10 mins, you can build a system that spots your best work hours, bad break patterns, focus drops, and the task order that actually helps you get more done.
2. This GPT-6 Astra Operating System is Absolute Cinema! Here're 5 Focused Levels to Build. A powerful AI system comes from building the right layers. Discover the 5 levels that turn GPT-6 Astra into a more consistent, personalized, and capable assistant.
3. This Is How You Can Easily Get a REAL Film Look From Seedance 2.5 (Free Claude Skills). You don’t need a complicated filmmaking workflow to get more cinematic videos. These updated free Claude Skills help you create shots that feel closer to real film.
4. Copy a $1K+/Day YouTube Channel with AI from Scratch. Step-by-Step (Everything You Need). A practical guide to using AI to make money with story-driven YouTube videos, from host creation to final export. The goal is a repeatable process you can actually run yourself.
D-DAY: WE’RE LIVE TONIGHT
Today it’s GPT-6 Astra and Claude Fable 5.1.
Tomorrow, there’ll be another “best model.”
But every time AI gets better, you shouldn’t have to rebuild your projects, instructions, and workflows from scratch just to use it.
Alex and Eric will guide you how to move from Claude to ChatGPT (and vice versa) without losing all that context behind. If you’ve ever stayed with just one AI simply because “it already knows me,” don’t miss this one.
📅 Today, September 16
⏰ 10:00 PM ET / 7:00 PM PT
🔴 LIVE on Zoom · 100% FREE
P/S: Join us live on Zoom and we’ll send you the webinar notes, prompts, and resources after the session.
TODAY IN AI
AI HIGHLIGHTS
👋 GPT-5.5 officially has an expiration date: Oct 14. OpenAI is removing it across all plans, so Codex users should start switching models now. Thanks for everything, 5.5!
😬 Sam Altman just spelled out these two AI risks that worry him most (try taking a wild guess first?) & that AI labs need to earn public trust. His full post explains why.
✍️ Grok Imagine now lets you precisely edit text inside generated images, including font, size, color, and alignment (here’s the demo). I already know what to test next!
👀 Google has opened Claude to all of its engineers through its system, while Gemini still comes first. But apparently even Google sees value in keeping Anthropic close.
💳 Meta is expanding paid AI across its apps with Meta One, starting at $7.99/month and more usage. Creators and businesses also get new tiers reaching $499/month.
⚡ AI data centers are getting ridiculously hungry for energy. By 2035, U.S. facilities could burn more natural gas than Germany and Japan combined (per this report)
💰 Big AI Acquisition: OpenAI bought Glass Imaging for over $300M. The startup, founded by ex-Apple engineers behind Portrait Mode, uses AI to make smartphone photos look much sharper and closer to DSLR quality.
NEW EMPOWERED AI TOOLS
🤖 Odyssey-3 learns how the world works from visual input, then controls robots, drones, and even video games like a real Physics Agent.
💳 tiun gives AI builders auth, billing, payments, customer data, and analytics in one system, so you can launch paid products from day one.
📱 Kilo Code lets you run coding agents, control VS Code and CLI sessions, review PRs, and manage your coding work from your phone.
🧠 siift turns your messy AI chats, ideas, and tools into a living business map, helping you connect decisions, evidence, risks, and next steps.
AI BREAKTHROUGH
The Shanghai AI Laboratory has released Atria Dawn Preview, an open-weight agentic model built for research and engineering workflows. It supports a 256K context window. What makes it interesting:
Atria is designed to move from a research question to an actual verifiable and reproducible result.
It can search for evidence, write and execute code, run experiments, inspect results, recover from failures, and produce reports or other finished outputs.
The lab groups its capabilities into Discovery, Creation, Delivery, and Cybersecurity.
Both the full model and an FP8 version have been released under an MIT license.
Atria scored 96.0 on DeepSearchQA, 92.5 on BrowseComp, 77.0 on BFCL v4, and 53.8 on AutomationBench. In those published comparisons, it beat or closely matched models including Kimi K3, GPT-5.6 Sol, and Opus 5 on some tasks.
The accompanying research paper describes Atria as being trained through verified tool interactions, with the goal of shifting AI from answering research questions toward helping complete whole research projects.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply