Google DeepMind's Gemini 4 Carbon launch

Google DeepMind released Gemini 4 Carbon, a model built around finding, verifying, and patching software vulnerabilities. It ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, and posts leading scores on security benchmarks: 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. Because of its exploit capabilities, access follows a staged rollout beginning with a subset of cyber defenders through the Fairwind Program, then trusted partners, developers, businesses, and consumers. HuggingFace has since released CWE-bench v2, an updated version of one of the benchmarks used to evaluate Carbon, adding broader vulnerability coverage, quality improvements, and greater transparency in its construction. Carbon also scores 96.2% on the GraphWalks long-context benchmark across 256K to 1M tokens, compared to Argon at 84.2%, GPT-6 Astra at 71.8%, and Claude Opus 5.5 at 66.8%. The Chrome team reported Carbon performed more than 10X better than another lab's large frontier model at generating high-quality patches. On safety, Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0% with attempts falling to 0.4%, cut high-severity misalignment alerts by nearly 80% in live coding environments, and achieved a 0.0% attack success rate on the Gray Swan Indirect Prompt Injection Benchmark, compared to Argon's 0.7%. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with prices set to double to $4 and $20.