HudHud Tech — Tech & AI news platform
Thursday, 29 October 2026
Breaking
Artificial Intelligence

Google Unveils Gemini 4 Carbon, Enhancing AI Capabilities

29 October 2026 2 min read هدهد تك | Hudhud Tech

Google has announced Gemini 4 Carbon, an upgraded version of its previous AI model, Gemini 4 Argon. This new model is designed to enhance capabilities in coding, agentic tasks, and scientific analysis, although it is not yet available for public use.

Gemini 4 Carbon reportedly excels in cybersecurity applications, showing a “significant leap” in vulnerability detection, validation, and remediation. According to Google, the model is twice as likely to generate code free of vulnerabilities compared to Argon and is ten times more effective at producing high-quality patches than non-Google models. Additionally, Carbon has improved its score by 89 points on WebDev Arena, now ranking fourth on the leaderboard.

While Google positions Carbon as a new model, it shares many similarities with Argon, performing slightly better in various tests. The model also surpasses competitors like GPT-6 Astra and Claude Opus 5.5 in most evaluations. However, the differences in benchmark scores are minimal, indicating that all frontier models perform closely.

Google claims that Carbon is the most capable frontier AI model available, with enhanced safeguards against prompt-injection attacks. In the Grey Swan evaluation, Carbon achieved a 0 percent success rate for such attacks, outperforming Argon and other models.

The rollout of Gemini 4 Carbon will follow a deliberate approach similar to that of Argon. Google is adhering to a voluntary testing framework established by the US government and plans to introduce the model through its Fairwinds cyber defense program. No specific timeline for a wide release has been provided, but if it mirrors Argon’s schedule, it could be available in a few weeks.

Initially, Carbon is expected to be accessible to AI Ultra subscribers and developers using the Gemini API. Google will offer a temporary discounted API rate of $2 per million input tokens and $10 per million output tokens, although the price will double after the introductory period. The end date for this discount has not been specified.

Related link: blog.google

 

Source: Ars Technica AI

What do you think of this story?

Share this story: