25launches

Gemini 4 Carbon

Google model for long agentic tasks and cyber defense

0 followers

Visit blog.google (opens in a new tab)

Artificial IntelligenceAPIDeveloper Tools

Gemini 4 Carbon is the model Google DeepMind calls its strongest, made for long agentic tasks and security work. Google says it can run for days on hard engineering problems, and reports 88.2% on SEC-bench Pro (finding flaws), 54.8% on ExploitGym and 71% on CWE-bench (patching). Paid API in AI Studio at $2/$10 per million input/output tokens until Nov 30, then $4/$20. It's in the Gemini app for AI Ultra subscribers. The full cyber version stays limited to vetted defenders.