Science & technology

Google Launches Gemini 4 Carbon

Is this AI launch a responsible leap in power and safety or hype dressed up with cherry-picked benchmarks?

2 hours ago·Updated 1 hour ago

Story

Above: The Google Gemini logo is seen displayed on a smartphone screen on Sept. 18. Image credit: Thomas Fuller/SOPA Images/LightRocket/Getty Images

The Facts

Google unveiled Gemini 4 Carbon on Wednesday, two weeks after releasing Gemini 4 Argon, describing it as its strongest frontier system with state-of-the-art cybersecurity and long-horizon capabilities, plus gains in coding and agentic tasks.
Reuters
Google
9to5Google
The system was announced hours before Alphabet reported third-quarter revenue of $127.6 billion, below analyst estimates of $128.2 billion. Shares fell more than 10% in extended trading, and CEO Sundar Pichai told analysts Google was again leading at the frontier.
Reuters
New York Post
Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, one point ahead of Claude Opus 5.5, according to the independent evaluator. Scores cannot be compared directly with the index's previous version.
Tech Times
Mashable
On cybersecurity benchmarks that Google ran without safety guardrails, Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71% on CWE-bench for patching at the root cause.
Google
THE DECODER
Anews
Google said Carbon produces secure code at more than twice Argon's internal success rate, and its Chrome team reported it performed over 10 times better at writing patches than a rival frontier system. Carbon gained 89 Elo points on WebDev Arena, reaching fourth.
Ars Technica
Tech Times
Engadget
Initial access starts this week, before expanding to other trusted partners, developers, businesses and consumers through the Fairwind Program. Introductory pricing is $2 per million input tokens and $10 per million output tokens.
India Today
9to5Google

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Pro-industry narrative

Pushing the frontier while locking down safety at the same time is exactly how AI should be built. Putting the strongest cyber defense tool in the hands of defenders first, before anyone else, is the only responsible call for a frontier system's release. Rapid-fire progress paired with voluntary government testing proves speed and caution can coexist.

Google

The Register

Industry-critical narrative

A shiny new system name doesn't mean much when the gains over the last version are razor thin and every rival lands in the same neighborhood. Company-picked benchmarks make a lousy substitute for letting real users test the thing. And a system nobody can actually use, with a fuzzy timeline and a discount that won't last, looks more like hype than a launch.

Ars Technica