Google DeepMind has unveiled Gemini 4 Carbon, a new frontier model designed for advanced cybersecurity defense and complex coding tasks. This model reportedly outperforms its predecessor, Gemini 4 Argon, and other leading models like GPT-6 Astra and Claude Opus 5.5 on several key benchmarks, particularly in vulnerability finding, exploitation, and patching. However, access to Gemini 4 Carbon is currently restricted to a select group of cybersecurity professionals and trusted partners, with no public API or open-weight release planned. AI
IMPACT Sets new SOTA on cybersecurity and coding benchmarks, but access is restricted to select partners.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Agents' Last Exam v1.1
- ExploitGym
- Fable 5.1
- FrontierSWE
- Gemini 4 Argon
- Google DeepMind
- GPT -6 Astra
- GraphWalks
- Opus 5.5
- RiemannBench
- SEC-bench Pro
- SWE-sweep
- Vending-Bench 2
- WebDev Arena
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →