Google targets cybersecurity and coding with Gemini 4 Carbon

Google DeepMind’s Gemini 4 Carbon is designed for vulnerability detection, secure code generation, scientific reasoning and autonomous engineering workflows. Google says the model improves on Gemini 4 Argon, including tasks that require sustained work over several days, such as tracing bugs and developing fixes across large software systems.

Google reports scores of 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation assessment and 71.0% on CWE-bench for vulnerability remediation. The company also says Carbon achieves more than twice the success rate of its predecessor in secure code generation on internal benchmarks. These evaluations indicate technical gains, but do not guarantee equivalent performance in real-world deployments.

Safety improvements include stronger alignment and resistance to malicious instructions. Google reports a 0.0% attack success rate on the evaluated Gray Swan Indirect Prompt Injection benchmark and approximately 50% fewer unsupported claims than Gemini 4 Argon in internal testing.

Initial access is restricted to selected cybersecurity defenders and trusted partners through Google’s Fairwind Program. Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with a 95% discount for cached input tokens. Prices will subsequently rise to $4 per million input tokens and $20 per million output tokens. Broader access for developers, businesses and consumers is planned, but Google has not confirmed a public release date.