Google DeepMind unveils Gemini 4 Carbon as its strongest frontier model with state-of-the-art cybersecurity and long-horizon capabilities
Gemini 4 Carbon leads on cybersecurity benchmarks, including SEC-bench Pro, ExploitGym, and CWE-bench, and introduces sustained agentic workflows for complex engineering tasks.
1 source · cross-referenced
- Gemini 4 Carbon is Google DeepMind’s strongest frontier model, emphasizing state-of-the-art cybersecurity and long-horizon capabilities.
Google DeepMind announced Gemini 4 Carbon, positioning it as its strongest frontier model to date. The model is designed to excel in cybersecurity and long-horizon agentic tasks, including sustained multi-step engineering workflows.
Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, reflecting significant gains in coding, agentic tasks, and cybersecurity-critical capabilities. On external cybersecurity benchmarks, it achieves 88.2% on SEC-bench Pro for vulnerability detection, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. Internally, Carbon delivers more than a twofold improvement in generating secure code compared to Argon, and in live coding environments, it shows reduced misalignment and fabrication rates.
The model is built to tackle complex, long-horizon engineering problems, with internal teams observing Carbon persist for days to root-cause and patch subtle bugs across large codebases and low-level systems. In web development, Carbon improves by 89 Elo points to reach #4 (1768) on WebDev Arena, reflecting enhanced performance in full-stack web development and multimodal understanding.
Gemini 4 Carbon is being rolled out through a phased approach, starting with a subset of cyber defenders this week. Access will expand to additional trusted partners via the Fairwind Program, with plans to make it available to developers, businesses, and consumers through Google’s product surfaces in the near future. The model will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at a 95% discount.
Gemini 4 Carbon builds on the multi-layered safety and security stack introduced with Argon, incorporating model training safeguards, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring. Google DeepMind reports continued progress in model alignment and safety, with Carbon extending these efforts further.
- Nov 11, 2026 · The Verge — AI
Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Trust79 - Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79