Google DeepMind has announced two new frontier AI models: Gemini 4 Carbon and Genie 4. Gemini 4 Carbon is designed for complex, long-horizon tasks, excelling in multi-day coding, cybersecurity defense, and bug patching, achieving top scores on several benchmarks. Genie 4 is a general world model capable of real-time, interactive simulations with realistic visuals and native audio generation, leveraging Gemini models for its core functionality. Both models are being rolled out to trusted testers and partners, with broader access planned for developers and enterprise users. AI
IMPACT These models signal a shift towards AI agents capable of complex, long-term autonomous tasks in coding, cybersecurity, and interactive simulation.
RANK_REASON Direct announcement from a frontier lab (Google DeepMind) of new models with specific capabilities and benchmark results. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
- Artificial Analysis Intelligence Index v5.0
- Chrome security team
- CWE-bench
- ExploitGym
- Fairwind Program
- Gemini 4 Argon
- Gemini Omni
- Google DeepMind
- Gray Swan
- SEC-bench Pro
- WebDev Arena
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →