Google Announces Gemini 4 Carbon

Google DeepMind announced Gemini 4 Carbon on October 28 as its strongest model, highlighting cybersecurity and long-horizon capabilities. The company says Carbon ranks first on the recently updated v5.0 Artificial Analysis Intelligence Index with a score of 59. It also reports results of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. The company says Carbon has reduced misalignment and fabrication rates in internal live coding environments.

Carbon follows Gemini 4 Argon, launched two weeks earlier. Google says Carbon is designed for sustained agentic workflows and that internal teams have used it for days to work through dead ends on complex engineering tasks, including finding root causes and patching subtle bugs in large codebases and low-level systems. It also improved across coding, agentic and scientific tasks, full-stack web development, and multimodal understanding. On WebDev Arena, Carbon rose 89 Elo points to rank fourth with a score of 1768.

Google reports that Carbon more than doubled Argon’s success rate for generating secure code on internal benchmarks. The Chrome team said Carbon performed more than 10 times better than another lab’s large frontier model at generating high-quality patches. Wiz also reported that Carbon led benchmarks for web penetration testing. The external cybersecurity benchmark results were evaluated without guardrails.

For safety, Carbon uses Argon’s layered protections, including model training, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring. Google says successful cheating on Terminal-Bench 4.0 fell to 0.0%, while cheating attempts fell to 0.4%; high-severity misalignment alerts in internal live coding environments dropped by almost 80%. Internal testing found approximately 2x fewer unsupported claims than Argon. Carbon also reduced overreach by 26% and under-service by 29%, and recorded a 0.0% attack success rate on Gray Swan’s Indirect Prompt Injection benchmark, compared with Argon’s 0.7%.

Google is beginning a limited rollout this week to a subset of cyber defenders, then plans to extend access gradually to trusted partners through its Fairwind Program. The company is participating in the U.S. government’s voluntary pre-release model testing process and says it will continue gathering feedback and refining safeguards before offering Carbon to developers, businesses, and consumers. Introductory prices match Argon’s: $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% below the input price. After the introductory period, prices will be $4 per 1M input tokens and $20 per 1M output tokens.

Google DeepMind announced Gemini 4 Carbon on October 28 as its strongest model, highlighting cybersecurity and long-horizon capabilities. The company says Carbon ranks first on the recently updated v5.0 Artificial Analysis Intelligence Index with a score of 59. It also reports results of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. The company says Carbon has reduced misalignment and fabrication rates in internal live coding environments.

Carbon follows Gemini 4 Argon, launched two weeks earlier. Google says Carbon is designed for sustained agentic workflows and that internal teams have used it for days to work through dead ends on complex engineering tasks, including finding root causes and patching subtle bugs in large codebases and low-level systems. It also improved across coding, agentic and scientific tasks, full-stack web development, and multimodal understanding. On WebDev Arena, Carbon rose 89 Elo points to rank fourth with a score of 1768.

Google reports that Carbon more than doubled Argon’s success rate for generating secure code on internal benchmarks. The Chrome team said Carbon performed more than 10 times better than another lab’s large frontier model at generating high-quality patches. Wiz also reported that Carbon led benchmarks for web penetration testing. The external cybersecurity benchmark results were evaluated without guardrails.

For safety, Carbon uses Argon’s layered protections, including model training, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring. Google says successful cheating on Terminal-Bench 4.0 fell to 0.0%, while cheating attempts fell to 0.4%; high-severity misalignment alerts in internal live coding environments dropped by almost 80%. Internal testing found approximately 2x fewer unsupported claims than Argon. Carbon also reduced overreach by 26% and under-service by 29%, and recorded a 0.0% attack success rate on Gray Swan’s Indirect Prompt Injection benchmark, compared with Argon’s 0.7%.

Google is beginning a limited rollout this week to a subset of cyber defenders, then plans to extend access gradually to trusted partners through its Fairwind Program. The company is participating in the U.S. government’s voluntary pre-release model testing process and says it will continue gathering feedback and refining safeguards before offering Carbon to developers, businesses, and consumers. Introductory prices match Argon’s: $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% below the input price. After the introductory period, prices will be $4 per 1M input tokens and $20 per 1M output tokens.