Google's Gemini 4 Carbon Guards the Gates as Nvidia Learns to Listen

Google's Gemini 4 Carbon Guards the Gates as Nvidia Learns to Listen

Google gives cyber defenders first dibs on its strongest model, OpenAI moves deeper into ServiceNow, and Nvidia ships a free transcriber that knows who said what.

Ledwyn Mena
Ledwyn Mena
Oct 29, 20263 min read

Grab your hoodie and your headphones, because today's AI news is half hacker movie, half podcast studio.

Gemini 4 Carbon Gets a Security Clearance

Google DeepMind announced Gemini 4 Carbon, just two weeks after Gemini 4 Argon showed up. 🛡️ Google calls it its strongest model yet for cybersecurity and long, multi-step jobs, and it ranks #1 on the Artificial Analysis Intelligence Index with a score of 59.

  • 🧩 On Google's own tests it scored 88.2% at finding software holes and 71.0% at patching them.
  • 🚪 For now, only a small group of cyber defenders gets access through Google's Fairwind Program, while it goes through the U.S. government's voluntary pre-release testing.
  • 🧾 Launch price matches Argon: $2 per million input tokens and $10 per million output tokens.

It's like a new superhero who has to pass a background check before the public gets to meet them.

OpenAI Moves Into ServiceNow's Office

OpenAI and ServiceNow expanded their partnership. 💼 A new ServiceNow plugin for ChatGPT lets IT teams find similar incidents, check how risky a change is and update records without leaving the chat, all while following their company's existing rules and permissions.

  • 🖥️ ServiceNow plans to give every employee ChatGPT Work and Codex.
  • 🔐 A joint security tool is set for early access in January 2027, with voice features expected early next year too.

Basically, the help desk just got a chat window that can actually press the buttons.

Nvidia's Free Transcriber Knows Who Said What

Nvidia released Nemotron 3.5 Transcribe, a free, open-weight model that turns recordings into text in 40 languages. 🎙️ It can translate 31 languages to or from English, handle recordings up to an hour long and tell up to eight speakers apart.

  • 📝 Nvidia says it's the top open-weight model on Hugging Face's Open ASR Leaderboard for English.
  • 🧩 It also released Granary v2, a giant open speech dataset for training.
  • ⏸️ One catch, per TechCrunch: it works on recorded audio, not livestreams.

No more group-project meeting notes that just say "someone said something about the deadline."

Other Cool Things to Know Today

  • 📰 Microsoft said its business with OpenAI brought in about $10 billion last quarter, roughly a tenth of its total sales.
  • 👾 OpenAI now lets some U.S. ChatGPT Pro users text their Dot agent over iMessage or RCS, in preview.
  • 🔥 Anthropic launched smart reports, surveys and pooled team budgets so Claude Enterprise customers can see what their AI spending actually buys.

For more such cool updates, follow us on Twitter!

Related reading