AI news · Google · Gemini 4 Carbon
Google’s Gemini 4 Carbon goes to cyber defenders first
Google says its new top model finds and fixes software flaws, and it leads an independent ranking. For now only vetted security firms can use it.
Google announced Gemini 4 Carbon on 28 October 2026 and calls it its strongest model (Google’s post). Google says Gemini 4 Argon came out two weeks earlier. Carbon’s pitch is security work: Google says it finds real software flaws, checks which can be exploited and writes the fix. It scores 88.2 per cent on SEC-bench Pro, a benchmark for spotting vulnerabilities. By the Chrome team’s measure, Carbon’s patches beat another lab’s large model more than tenfold, Google says.
For now access stops at a hand-picked set of security companies, The Verge reports. Google says it will let in more trusted partners gradually through its Fairwind programme, and developers, businesses and consumers later, with no date. On the Artificial Analysis Intelligence Index, an independent ranking now in version 5, Carbon scores 59 and places first, with Claude Opus 5.5 and GPT-6 Astra behind.
Google gives Carbon Argon’s introductory price: $2 per million input tokens and $10 per million output tokens, tokens being the chunks of text a model reads and writes. When that period ends, the price rises to $4 and $20.
None of the four Google AI plans in our 28 October price check names a Gemini 4 model. The Free card lists Gemini 3.6 Flash and some access to 3.1 Pro, and AI Pro costs $19.99 a month.
With Carbon held inside Fairwind and no Gemini 4 model named on any of the four plan cards we read, paying $19.99 a month for AI Pro does not buy Carbon today; businesses reach it only when Google opens it to them, on no stated date.
Sources
- Google: Introducing Gemini 4 Carbon. Checked
- The Verge: Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity. Checked