As we speak, we’re asserting the subsequent frontier mannequin: Gemini 4 Carbon, our strongest mannequin with state-of-the-art cybersecurity and lengthy horizon capabilities.
Carbon ranks #1 on the just lately up to date v5.0 of Artificial Analysis Intelligence Index at 59. It exhibits important beneficial properties throughout coding and agentic duties in addition to throughout key capabilities vital for cybersecurity protection, touchdown a agency lead in SEC-bench Professional (88.2%) for locating vulnerabilities, ExploitGym (54.8%) for exploitation, and CWE-bench (71.0%) for patching. On inner benchmarks, Carbon drives a powerful uplift in producing safe code, and in dwell coding environments with our personal groups it exhibits decreased misalignment and fabrication charges.
Constructing on Gemini 4 Argon, which was launched two weeks in the past, Carbon illustrates the tempo at which we’re pushing the frontier ahead. Our researchers and fashions are working hand in hand to advance capabilities quicker, whereas strengthening our safeguards in tandem.
We’re taking a deliberate method to rollout and are actively engaged within the U.S. authorities’s voluntary course of for pre-release mannequin testing. Beginning this week, we’ll make Carbon accessible to a subset of cyber defenders and can progressively section entry to extra trusted companions via our Fairwind Program. We’re persevering with to collect suggestions and iterate on guardrails earlier than making Carbon accessible to builders, companies, and shoppers through our product surfaces as quickly as doable.
Gemini 4 Carbon will launch on the similar introductory worth
as Argon at $2 per million enter tokens and $10 per million output tokens, with cached enter tokens priced at 95% off enter token worth.
