Gemini 4 Carbon tops Intelligence Index v5 but trails Claude Opus 5.5 in coding

Gemini 4 Carbon builds on Gemini 4 Argon, which Google released just two weeks earlier on October 14, and carries over the same multi-layered safety measures. Google describes the new model as its strongest yet, designed for long-running engineering tasks and cybersecurity work, and says internal teams have seen it persist through multi-step engineering problems over several days — a claim the company has not independently verified.

The ranking picture is more nuanced than the headline number suggests. On the Intelligence Index v5, Carbon takes first place overall with 59 points, but it tops only four of the nine underlying evaluations. Artificial Analysis credits its lead to complex reasoning, knowledge, and science tasks, while noting the model is not top in coding or knowledge work. On the separate Coding Agent Index, Carbon sits 3 points behind Claude Opus 5.5 running in Claude Code at 62 points, though the evaluator frames this as a notable improvement for Google, which has lagged in coding.

Pricing and access remain open questions. The introductory rates mirror Argon's, but Google has not disclosed how long they last, and the model is initially limited to a subset of cybersecurity defenders before broader distribution through the Fairwind Program. For teams weighing a switch, the practical takeaway is to map current use cases and budget for both pricing tiers while waiting for a firm general-availability date.