Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Gemini 4 Carbon took the top spot on the Artificial Analysis Intelligence Index (v5.0), ahead of Claude Opus 5.5 and GPT-6 Astra, and outperformed leading OpenAI and Anthropic models on some benchmarks.
- Google called Carbon a "significant leap" in cyberdefense, an improvement over Argon, the model it released earlier this month.
- SEC-bench Pro, which measures real-world bug discovery, ranks Carbon No. 1, and Carbon also leads ExploitGym, the benchmark testing whether a model can turn a vulnerability into an exploit.
- Fairwind Program access starts narrow: only a select group of cybersecurity defense companies can use Carbon now, with other partners to follow.
- Carbon is built for long-horizon work — Google's internal teams have seen it run for days through dead ends on complex, multi-step engineering tasks — while resisting prompt injection, avoiding shortcuts and collaborating with users, according to the company.
- Argon remains the baseline on price: Carbon launches at the same introductory rate of $2 per million input tokens and $10 per million output tokens, while cutting unsupported claims two times versus Argon.
Why it matters: Cybersecurity defense companies are the first constituency Google names for Carbon, and they get the model at $2 per million input tokens and $10 per million output tokens — the same introductory price as Argon. For that narrow group, Google's top-ranked bug-hunting and exploit-generation scores become usable tooling before wider Fairwind Program access.
Ask SkimNews
