Skip to content
Computing / NEWS

Google announces Gemini 4 Carbon, a staged cybersecurity-focused frontier model

Google announced Gemini 4 Carbon two weeks after Gemini 4 Argon, citing security and coding gains and a staged rollout under US testing and Fairwinds.

Sunlight enters a forest in Bad Pyrmont, Germany
Context photograph · Sebastian Unrau / Unsplash
THE KEY FINDING

The story in focus

Ars Technica reports Google announced Gemini 4 Carbon on October 28, 2026, two weeks after Gemini 4 Argon. Google cites gains in vulnerability detection and patching, slim benchmark leads over rival models, and a staged rollout under US voluntary testing and the Fairwinds cyber defense program, without a wide-release timeline.

  • Google announced Gemini 4 Carbon on October 28, 2026, two weeks after rolling out Gemini 4 Argon.
  • Google says Carbon is twice as likely to produce vulnerability-free code as Argon and improved by 89 points on WebDev Arena.
  • Google claims Carbon beat GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 in almost every test, though Ars Technica notes gaps of only a few points.
  • Carbon resisted every attempted prompt-injection attack in the Grey Swan evaluation, a 0 percent score, according to Google.
  • Google is gating the release behind US voluntary testing and the Fairwinds cyber defense program; no wide-availability date was given.

What Google announced

Google announced Gemini 4 Carbon on October 28, 2026, two weeks after the release of Gemini 4 Argon, Ars Technica reported. Google describes Carbon as an improvement on Argon rather than a smaller Flash variant. Ars Technica notes the two models share much in common and that Carbon could be a post-training upgrade, a possibility the article raises without confirmation.

Google says the model makes a significant leap in vulnerability detection, validation and remediation. According to Google, Carbon is twice as likely as Argon to generate code free of vulnerabilities, and the Chrome team claims it is 10 times better than non-Google models at producing high-quality patches. Google also says Carbon gained 89 points on WebDev Arena to rank fourth.

Benchmarks and release plans

Google positions Carbon as the most capable frontier model. Ars Technica reports Carbon edges Argon by slim margins on every test Google supplied and beats GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1 in almost every case, though the article notes all frontier models score in the same neighborhood, with gaps of a few points. In Google's Grey Swan prompt-injection evaluation, Carbon scored 0 percent, versus 0.7 percent for Argon, 1 percent for Opus 5.5 and 8.5 percent for GPT-6 Astra.

Google is not setting a wide-release date. It says it is working within the US government's voluntary testing framework and plans to roll Carbon out through its Fairwinds cyber defense program. The article says Carbon will probably reach AI Ultra subscribers and Gemini API developers first, with introductory pricing of $2 per million input tokens and $10 per million output tokens that will later double; Google declined to give an end date for the discount.

When did Google announce Gemini 4 Carbon, and how is its release being staged?

Ars Technica reports Google announced it on October 28, 2026, two weeks after Gemini 4 Argon. Google is gating wider availability behind US voluntary testing and its Fairwinds cyber defense program and gave no wide-release timeline.

Sources & further reading

Source report: 28 Oct 2026. Event and publication dates may differ.

  1. Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use ↗Ars Technica

Help keep the record accurate. Send a correction.