Overview
- Google has officially launched Gemini 4 Carbon, a new frontier large model that the company positions as its most capable for long, multi-step engineering and security workflows.
- Google published benchmark results it says show leading performance on industry tests, including 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploit verification and 71.0% on CWE-bench for patch repair.
- The company reported internal and partner tests that show large gains over the prior Gemini 4 Argon release, including more than twice the success rate on secure code generation and Chrome team results it describes as roughly ten times better for high-quality patches.
- Google says Carbon includes safety and alignment engineering that cuts hallucinations and cheating in its tests, refuses unsafe operations, flags missing evidence and reduces injection-style failures in its internal evaluations.
- Access will roll out in stages: Google is starting with a small group of vetted cybersecurity defenders as part of coordinated voluntary U.S. government testing and will widen availability through its Fairwind partner program before opening to developers, enterprises and consumers; introductory token pricing has been published.