Google Unveils Gemini 4 Carbon, Touting Gains in Coding, Agents and Cybersecurity

TL;DR: Google has launched Gemini 4 Carbon, its newest frontier model, citing "significant gains" in coding, agentic tasks and cybersecurity.

Key points

• In a Wednesday announcement, Google said Gemini 4 Carbon improves markedly in coding, agentic tasks and cybersecurity.
• Google highlighted the model's first-place result on the Artificial Analysis Intelligence Index (v5.0).
• Claude Opus 5.5 and GPT-6 Astra trail Carbon on that benchmark.

Why it matters

Frontier-model competition is shifting from general capability toward practical areas such as coding, agents and security. The ranking comes from one benchmark that Google chose to emphasize, so real-world performance in your own workflows is worth testing separately.

Read more

• Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity — The Verge AI