Google launches Gemini 4 Carbon with focus on coding and cybersecurity
publishers 15articles 15first reported 28 Oct, 20:00 UTCdeveloping since 29 Oct · 4 editions
Google launched Gemini 4 Carbon on Wednesday, describing the frontier AI model as an advance in coding, agentic tasks and cybersecurity over Gemini 4 Argon, which it released earlier this month.
According to benchmarks published by Google, Carbon scored 59 on the Artificial Analysis Intelligence Index v5.0, placing it narrowly ahead of Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra. Google also said Carbon ranks first on SEC-bench Pro, which assesses models’ ability to find real-world software bugs, and ExploitGym, which measures exploitation of vulnerabilities.
The company said Carbon is better at finding and prioritizing vulnerabilities, addressing their root causes and generating secure code. Google claims it is twice as likely as Argon to produce secure code and is more resistant to prompt-injection attacks and attempts to exploit shortcuts. It also said the model can sustain complex engineering work for days and makes half as many unsupported claims as Argon.
Carbon is initially available only to a select group of cybersecurity defense companies for testing. Google plans to extend access to additional trusted partners through its Fairwind Program before a broader release. The Verge reported that Google will offer the model at an introductory price of $2 per million input tokens and $10 per million output tokens.
HOW THIS STORY WAS MADE
Written from 4 articles, 2 with the publisher's own text; 4 independent newsrooms once syndicated copies count as one; this version written 12 h after the record first saw the story; the editor kept it as written.
- Publishers
- 15
- Source articles
- 15
- Given to the writer
- 4, 2 with the article's own text
- Independent newsrooms
- 4
- This version written
- 12 h after the record first saw the story
- Second model (editor)
- kept as written
- Publication gate
- passed
- Human review
- none
WHAT THE EDITOR DID
The second model checked the writer's draft against the same sources and published it as written.
Written by a language model from the sources above, then checked by a second model that may only cut, attribute or correct. How it works →