# Google DeepMind Launches Latest Flagship Large Model Gemini 4 Carbon

- Category: Flash (快讯)
- Published: 2026-10-29T11:58:00.000Z
- Source date: 2026-10-29
- HTML: https://www.88z.ai/en/news/flash-20261029-e3d28fd89a38fdd2
- Markdown: https://www.88z.ai/en/news/flash-20261029-e3d28fd89a38fdd2.md
- Other language: https://www.88z.ai/news/flash-20261029-e3d28fd89a38fdd2.md

Google DeepMind has released its flagship model, Gemini 4 Carbon, which topped AI intelligence benchmarks and surpassed key competitors. The model features long-term processing capabilities for large codebases, focusing on security vulnerability detection and repair, and is currently available only to select cybersecurity professionals.

## At a glance

- Gemini 4 Carbon topped the AI Intelligence Index with a score of 59, surpassing Claude Opus 5.5 and GPT-6 Astra.
- The model supports continuous multi-day processing of large codebases for security repairs and is currently available only to select cybersecurity personnel.

## The update

Google DeepMind has officially launched its latest flagship large model, Gemini 4 Carbon. In the Artificial Analysis Intelligence Index v5.0 evaluation, the model ranked first with a score of 59, outperforming mainstream models such as Claude Opus 5.5 and GPT-6 Astra. Gemini 4 Carbon demonstrates strong engineering capabilities, able to process large codebases continuously for several days, covering the entire workflow from vulnerability discovery and exploit verification to final repair. Furthermore, the model leads in multiple security benchmarks and shows significantly improved system reliability. Currently, Gemini 4 Carbon is available this week only to select cybersecurity defense personnel, with the specific release time for general users yet to be determined.

## Sources

- [APPSO](https://mp.weixin.qq.com/s/tJZZvtx5-yOavPwcxOfPiw) (2026-10-29)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.