Google has announced Gemini 4 Carbon, its latest frontier model, with a particular focus on cybersecurity and complex, long-running tasks. According to Google's announcement, it also improves on Gemini 4 Argon in coding and agentic work.
Cybersecurity and benchmarks
Google reports results of 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for validating exploitability, and 71.0% on CWE-bench for patching. The company says these external benchmark evaluations were conducted without guardrails. They are reported test results, not a guarantee that the model will find or fix every real-world vulnerability.
In the newly updated Artificial Analysis Intelligence Index v5, Carbon with high reasoning scored 59 and ranked first at the time of the announcement. The original article noted a one-point lead over Claude Opus. Artificial Analysis confirms Carbon's first place and reports it was second to Opus 5.5 on its separate Coding Agent Index. Rankings depend on the benchmark, model settings and date; a lead on one index does not mean a lead on every task.
Access and pricing
For now, access is limited. Google says it is beginning with a subset of cyber defenders and will gradually add trusted partners through its Fairwind Program, before offering Carbon to developers, businesses and consumers. It has not given a specific public release date, so we will have to wait and see.
Google says Carbon will launch at the same introductory API price as Argon: $2 per million input tokens and $10 per million output tokens. Cached input is discounted by 95%, or $0.10 per million at those introductory rates. After the introductory period, Google says the standard input and output prices will be $4 and $20 per million tokens. The announcement does not specify when Carbon's offer ends. API pricing is separate from consumer subscription access.
Translated from the original Chinese article, published October 28, 2026. Access, benchmark context and pricing checked against the sources linked above.
Comments NOTHING