Google's Gemini 4 Carbon rolls out to paying developers

Google positions Carbon as its strongest model, built for tasks that push the limits of model intelligence such as complex multi-step coding and open-ended scientific research. Its earlier Gemini 4 Argon, launched last month, remains the company's recommended choice for most software engineering and knowledge work. Carbon carries the same per-token price as Argon, but Google says it reasons more deeply and runs longer verification loops, so the total bill for a task is likely to be higher.

The most striking early result comes from Google DeepMind, which used agents running on Carbon, alongside Claude Opus 5.5, to formally verify core security guarantees in libsignal, the reference implementation of the Signal protocol also used in WhatsApp and Google Messages. Carbon-based agents generated roughly 6.5 million lines of Lean proofs, with five researchers completing the work in four weeks, which Google says has traditionally taken specialists years. The guarantees, including forward secrecy and post-compromise security, hold for the compiled machine code, though the results cover Google's own build of the code and exclude side channels. Signal CTO Ehren Kret said the work gives greater confidence in the code's privacy guarantees.

During the Fairwind trial, Google says security firm Wiz used Carbon to find zero-day vulnerabilities at two major SaaS providers within two hours, and Google's Cloud Vulnerability Research team used it to prove two open-source bugs were critically exploitable, prompting maintainers to reclassify and patch them. Google is also giving 10,000 scientists free access for six months and previewing a "swarm" feature in Antigravity that turns a single prompt into a coordinated team of agents.