Home › News › Technology › Google Gemini
News · TechnologyGoogle unveils Gemini 4 Carbon, its most powerful AI, built for coding and fending off hackers
Google announced its new artificial intelligence model on October 28, and it tops an independent ranking. For now, only some cybersecurity experts can use it.
Translated from the original Spanish article. Read in Spanish
News published on October 29, 2026. The situation may have changed since then.
On October 28, 2026, Google unveiled Gemini 4 Carbon, its new artificial intelligence model. The company describes it as its most powerful model so far.
It arrives just two weeks after Gemini 4 Argon, the previous model.
What it is for
Carbon is designed for long, complex tasks, especially coding and cybersecurity.
According to Google, its internal teams have seen it work for days on large code projects until it finds the cause of a bug and fixes it.
The numbers
- It ranks number 1 on the Artificial Analysis intelligence index, with 59 points. - It found security flaws in 88.2% of the SEC-bench Pro tests. - It climbed to 4th place on WebDev Arena, a ranking for building websites.
Compared with Argon, Google says it generates secure code with more than twice the success rate and makes half as many unsupported claims.
Google says that, in its tests, Carbon resisted every attempt to trick it with hidden instructions.
Who can use it
For now, it is not open to the public. It goes first to some cybersecurity experts and, little by little, to other trusted partners.
Google says it will reach developers, businesses and users soon, but it has not given a date. It is also taking part in the U.S. government's voluntary testing before launching models.
How much it will cost
For developers, the introductory price will be $2 per million input tokens and $10 for output, the same as Argon. Afterwards it will rise to $4 and $20.
Tokens are the chunks of text the AI processes.
More stories on video
This story is web-only. Every week on our social media we tell the stories behind brands and the news.
Does an AI that can find security flaws make you feel safer or more afraid?
Tell us in the comments on our social media.
Sources (in Spanish)
- Anuncio, capacidades, cifras, acceso y precios: Blog de Google (28 de octubre de 2026).
- Disponibilidad y comparación con Argon: 9to5Google (28 de octubre de 2026).
Photos are from Wikimedia Commons under free licenses; the author and license are shown under each image.
