Google targets cyber threats with Gemini 4 Carbon
The new model leads key security benchmarks and costs $10 per million output tokens.
By Fatima Davis ·
Google released Gemini 4 Carbon on Wednesday. The model currently ranks first on the Artificial Analysis Intelligence Index v5.0. It sits ahead of Claude Opus 5.5 and GPT-6 Astra in that specific ranking.
The company stated that Carbon provides better results in finding vulnerabilities and writing secure code. It holds the top position on SEC-bench Pro. This benchmark tests an AI's ability to spot real-world bugs. Carbon also leads ExploitGym, which measures how a model turns a vulnerability into an exploit.
Access to the model is limited for now. It is available to a select group of cybersecurity defense companies. Google plans to widen availability to other partners in its Fairwind Program. The new release improves upon Argon, a model Google launched earlier this month.
Google noted that internal teams observed Carbon running for days. This duration allowed it to push through dead ends in complex, multi-step engineering tasks. The model is designed to resist prompt injection attacks. It also avoids shortcuts when working with users.
The company claimed Carbon cuts unsupported claims by two times compared to Argon. This change aims to offer more transparency and honesty in outputs. Google positioned the release as a step forward in pushing the frontier of AI capabilities.
Pricing for Gemini 4 Carbon matches the introductory rate for Argon. Input tokens cost $2 per million. Output tokens cost $10 per million.
Source: The Verge
