Google says Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. The company reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them and 71.0% on CWE-bench for patching them, plus an 89-point Elo jump on WebDev Arena, where it ranks fourth. According to Google, in internal testing the model worked for days on multi-step engineering tasks such as root-causing and patching bugs across large codebases.
Google says the model more than doubles the success rate for generating secure code compared with Argon, and that the Chrome team reported patches more than ten times better than another lab's frontier model. On safety, the company says successful cheating on Terminal-Bench 4.0 fell to 0%, high-severity misalignment alerts in internal coding environments dropped by almost 80%, and no attack succeeded on Gray Swan's indirect prompt injection benchmark. According to Ars Technica, Carbon beats GPT-6 Astra and Claude Opus 5.5 in almost every test Google published, but the gaps are only a few points.
The rollout is staged: Google is taking part in the US government's voluntary pre-release model testing process, and access will widen gradually to more partners through its Fairwind program. The introductory price matches Argon's, $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Neither the date of availability for developers, businesses and consumers nor the end of the introductory pricing has been announced.
