Google releases Gemini 4 Carbon with expanded cybersecurity capabilities

Google announced Gemini 4 Carbon on Wednesday, October 29, 2026, its latest frontier AI model offering improvements in coding, agentic tasks, and cybersecurity. The model currently holds the top position on the Artificial Analysis Intelligence Index (v5.0) benchmark, placing ahead of Claude Opus 5.5 and GPT-6 Astra.

Carbon is an upgrade over Gemini 4 Argon, which Google released earlier in October 2026. For now, access is limited to a select group of cybersecurity defense companies, with a planned expansion to additional partners through Google’s Fairwind Program.

Google describes Carbon as delivering a “significant leap” in cyberdefense, citing its ability to find vulnerabilities and generate secure code. The model holds the number one position on two cybersecurity benchmarks: SEC-bench Pro, which measures an AI model’s ability to discover real-world bugs, and ExploitGym, which measures how a model can turn a vulnerability into an exploit. Carbon also outperformed GPT-6 Astra specifically on the SEC-bench Pro benchmark.

The model is designed for longer, more complex tasks. Google says its internal teams have observed Carbon running for days to work through multi-step engineering challenges. The company also says Carbon is better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users. Compared to Argon, Carbon is said to cut “unsupported claims” by half, which Google frames as an improvement in transparency and honesty.

Carbon will be priced at $2 per million input tokens and $10 per million output tokens — the same introductory pricing as Argon. For organizations working in cybersecurity defense, the model’s benchmark performance and extended task capabilities may make it a notable option, though access remains restricted while Google expands its partner program.