Google Launches Gemini 4 Carbon AI Model to Detect and Fix Software Security Flaws

UPDATE - follow-up to What is Gemini Carbon 4, Google's new AI model for finding, fixing security vulnerabilities?: Google rolled out its latest frontier AI model named Gemini 4 Carbon on Wednesday, October 28, to strengthen cybersecurity defence.

Performance and Benchmarks

Carbon is the company's second major model release in a single month following Gemini 4 Argon, bringing upgrades in coding, agentic tasks, and cybersecurity. Google reports that Carbon outperforms OpenAI's GPT-6 Astra along with Anthropic's Claude Opus 5.5 and Claude Fable 5.1 across nearly all tests. The model ranks number one on the v5.0 Artificial Analysis Intelligence Index with a score of 59. It also achieves 88.2 percent on SEC-bench Pro for finding vulnerabilities, 54.8 percent on ExploitGym for exploitation, 71.0 percent on CWE-bench for patching, and increased its score by 89 points on WebDev Arena to rank fourth on the leaderboard.

Capabilities and Pricing

Built for extended engineering tasks, Carbon can run for days to solve complex, multi-step actions without stopping at dead ends. Google states the model resists prompt injection attacks, avoids shortcuts, and collaborates directly with users. It cuts unsupported claims by two times and generates vulnerability-free code twice as often compared to Argon. Google Chrome engineers also found Carbon ten times better at generating high-quality patches than non-Google models. At launch, the model costs $2 per million input tokens and $10 per million output tokens, matching Argon's introductory pricing.

Rollout and Business Impact

Google has adopted a careful rollout approach by making Carbon available initially to select cybersecurity defense companies before expanding it to partners in its Fairwind Program. The model's debut coincided with Google's earnings report showing strong cloud computing growth but weak search advertising revenue, leading to a nearly 10 percent drop in stock price during after-hours trading.

Context

Major technology companies frequently release specialized artificial intelligence models to capture market share in enterprise software and cloud services. Security-focused AI tools are designed to help human engineers find and fix software bugs faster than traditional manual methods.