Google has announced Gemini 4 Carbon, its latest frontier model, highlighting capabilities in cybersecurity, coding and sustained, multi-step tasks. The company says Carbon ranks first on the updated Artificial Analysis Intelligence Index v5.0, with a score of 59.
On external cybersecurity benchmarks, Carbon scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for testing exploitability and 71.0% on CWE-bench for patching vulnerabilities. Google says these evaluations were conducted without guardrails. The company also reports that Carbon more than doubled Argon’s success rate at generating secure code on internal tests.
The model is designed for long-running engineering workflows. Google says internal teams have used it for tasks lasting days, including tracing and fixing subtle bugs across large codebases and low-level systems. It also reports improved results in web development: Carbon gained 89 Elo points to reach fourth place on WebDev Arena, with a score of 1,768.
Google says it has made changes to Carbon’s alignment and safety measures, including reducing unsupported claims and attempts to exploit shortcuts. In internal testing, successful cheating on Terminal-Bench 4.0 fell to 0.0%, while attempts dropped to 0.4%. The company also reports an almost 80% reduction in high-severity misalignment alerts in live coding environments, and a 0.0% attack success rate on the Gray Swan Indirect Prompt Injection benchmark.
Carbon uses the same multi-layered safety and security approach as Gemini 4 Argon, which Google says includes model training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring. Google is participating in the U.S. government’s voluntary pre-release model testing process and will begin providing access this week to a subset of cyber defenders. Access is expected to expand gradually to trusted partners through the Fairwind Program, with wider availability for developers, businesses and consumers planned later.
Gemini 4 Carbon will have the same introductory API pricing as Argon: $2 per million input tokens and $10 per million output tokens. Cached input tokens will cost 95% less than the standard input price.
Sources: Google DeepMind Blog.