Google ships Gemini 4 Carbon, its second flagship model in a month

Google announced Gemini 4 Carbon on October 28, its next frontier model and, according to The Information, the second major model it has shipped in a month. 9to5Google notes that it lands roughly two weeks after the previous release, Argon.

Google's own framing centres on gains in coding, agentic tasks and cybersecurity defence, specifically vulnerability discovery. The company says Carbon ranks first on the recently updated v5.0 of the Artificial Analysis Intelligence Index with a score of 59.

Coding and web development get particular attention. Google says it worked from user feedback after the last launch to improve coding, agentic and scientific tasks, full-stack web development and multimodal understanding. With better output quality and artifacts, Carbon gains 89 Elo points to reach fourth place on WebDev Arena at 1769.

Security is treated as a headline capability rather than a footnote. Google cites an 88.2% result on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching. Internal benchmarks reportedly show a stronger uplift in generating secure code, with lower misalignment and fabrication rates in live coding environments used by Google teams.

The alignment claims come with numbers too. Carbon cuts unsupported claims by roughly 2x compared with Argon, retrieves relevant sources and surfaces discrepancies to the user before acting when a request is ambiguous, and reduces overreach by 26% while cutting under-service by 29%, changes aimed at reliable end-to-end execution of multi-step plans.

Availability and pricing are set out plainly. Google says Carbon is coming soon to developers, businesses and consumers, with cyber defenders getting access first this week. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input at 95% off the input price; afterwards it moves to $4 and $20 per million tokens respectively.

Put the cadence together and the strategy becomes visible. Two flagship models in a month, each emphasising coding and agentic work, suggests Google's release cycle for this line has compressed from quarters to weeks.

The caveat is that every benchmark and safety figure above is Google's own, and independent reproduction will take time. The introductory pricing also reads as a customer-acquisition window; what gets tested next is stability and cost under real production load.

Three things to watch: when access widens beyond cyber defenders, whether outside evaluations corroborate the published numbers, and whether a model every two weeks is a pace Google can sustain.