New to this? Read it in simple words
- Google says Gemini 4 Carbon is its strongest AI model so far.
- Only some cyber defenders can use it this week.
- An independent tester ranks it first, with a score of 59.
- Claude Opus 5.5 is second with 58, so the lead is small.
- Cyber defender
- A person or team whose job is to protect computer systems from attacks.
- Intelligence Index
- Artificial Analysis’s combined score from many tests of what AI models can do.
- Token
- A small piece of text, about part of a word, that AI models read and write and charge for.
Carbon is Google’s second new frontier model in a month. Like Argon before it, it reaches cyber defenders before anyone else.
What Google released
Google calls Gemini 4 Carbon “our strongest model with state-of-the-art cybersecurity and long horizon capabilities.” Long horizon means the model can keep working on a long task with many steps.
“Starting this week, we will make Carbon available to a subset of cyber defenders,” Google wrote. More trusted partners will follow through its Fairwind Program. Developers, businesses and consumers will get it “soon,” Google says, without giving a date.
Carbon builds on Gemini 4 Argon, which Google announced on September 30 and opened to paying users on October 13. Google says Carbon uses the same safety and security systems as Argon.
From Google’s announcement and Artificial Analysis, October 28, 2026.
What the outside test found
Artificial Analysis scores Carbon 59 on version 5 of its Intelligence Index, first of 52 models. Anthropic’s Claude Opus 5.5 at its highest setting is second, with 58.
The index was renewed on the same day, so scores from the older version can’t be compared with the new ones, Mashable notes. Artificial Analysis lists no public API for Carbon yet.
Google’s own claims
Google says Carbon scored 88.2% on SEC-bench Pro, a cybersecurity test, with its guardrails switched off for the test. It also says Carbon writes secure code more than twice as often as Argon. These are Google’s own figures.
Google also says Carbon cheated on none of the tasks in the coding test Terminal-Bench 4.0. Serious alerts about misaligned behaviour fell by almost 80%, it says. No outside group has checked these numbers yet.
At launch, Carbon costs $2 per million input tokens and $10 per million output tokens, the same introductory price as Argon. Later it will cost $4 and $20, Google says. Google also says it takes part in the US government’s voluntary testing of models before release.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3
- 4Research · October 28, 2026Google Debuts New AI Models, Genie 4 and Gemini 4 Carbon: What You Need to Know CNET
- 5Research · October 28, 2026Gemini 4 Carbon is Google’s second new frontier model this month Android Authority
The ranking comes from Artificial Analysis’s own test. The cybersecurity and safety numbers are Google’s, and no outside group has checked them yet.
Google’s Gemini 4 Carbon tops a new independent test. Silicon AI News, October 28, 2026. https://siliconainews.com/stories/google-gemini-4-carbon
Google’s new frontier model ranks first on an independent test, though only cyber defenders get it now and the lead is one point. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.