In brief
Google announced its new Gemini 4 Carbon AI model on October 28, 2026, saying it tops the updated Artificial Analysis Intelligence Index and excels in coding and security benchmarks. The company plans a phased rollout beginning with cybersecurity defenders, followed by developers, businesses and consumers, and will offer tiered pricing that starts at $2 per million input tokens. In tests, Gemini 4 Carbon achieved 88.2% on SEC‑bench Pro and scored 89 points higher than its predecessor on the WebDev Arena leaderboard.
Google revealed its next‑generation AI model, Gemini 4 Carbon, on Wednesday, October 28, 2026, saying the system outperforms earlier versions on coding and cybersecurity tests. The tech giant reported that Gemini 4 Carbon ranks first on the updated version 5.0 of the Artificial Analysis Intelligence Index with a score of 59.
According to the company, the new model shows an 89‑point improvement to fourth place on the WebDev Arena leaderboard for full‑stack web development work, and it achieved high marks on security challenges: 88.2% on SEC‑bench Pro, 54.8% on ExploitGym and 71% on CWE‑bench for vulnerability detection, exploitation and patching.
Google noted that internal tests indicated Gemini 4 Carbon reduces misalignment and fabrication in live coding environments, cuts unsupported claims by roughly half compared with its Argon predecessor, and improves source retrieval. It also reports a 26% cut in task overreach and a 29% drop in under‑service during multi‑step workflows.
The company plans to release Gemini 4 Carbon first to cybersecurity defenders this week, before rolling it out to developers, businesses and consumers. Introductory pricing will be $2 per 1 million input tokens and $10 per 1 million output tokens, with a 95% discount on cached input tokens. Standard rates will later rise to $4 per input and $20 per output.
