Gemini 4 Carbon Tops AI Index, Leads Cyber Benchmarks — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google announced Gemini 4 Carbon as its next frontier model, taking the #1 spot on the recently updated v5.0 Artificial Analysis Intelligence Index with a score of 59, with gains across coding and agentic tasks.
- Gemini 4 Carbon posted state-of-the-art results on three external cyber benchmarks evaluated without guardrails: 88.2% on SEC-bench Pro for finding real-world vulnerabilities, 54.8% on ExploitGym for validating exploitability, and 71.0% on Collinear's CWE-bench for root-cause patching.
- Gemini 4 Carbon is built for sustained agentic work — Google's teams saw it run for days to root-cause and patch subtle bugs across large codebases and low-level systems — and it jumped +89 elo points to #4 (1768) on WebDev Arena.
- Chrome reported that Gemini 4 Carbon performed more than 10X better than another lab's large frontier model when generating high-quality patches; Carbon also more than doubled Argon's success rate at generating secure code on internal benchmarks, while Wiz observed it leading web penetration-testing benchmarks used in its Scan for Good initiative.
- Gemini 4 Carbon will launch at the same introductory price as Gemini 4 Argon, released two weeks ago: $2 per million input tokens, $10 per million output tokens, and cached input tokens priced at 95% off the input token price.
- Fairwind Program access will roll out in stages — a subset of cyber defenders starting this week, then additional trusted partners — while Google engages in the U.S. government's voluntary process for pre-release model testing and continues iterating on guardrails.
- Gemini 4 Carbon cut successful cheating on Terminal-Bench 4.0 to 0.0% with attempts to cheat dropping to 0.4%, reduced high-severity misalignment alerts by almost 80% in internal live coding environments, and reached a 0.0% attack success rate on Gray Swan's Indirect Prompt Injection benchmark versus Argon's 0.7%.
Why it matters: Cyber defenders get first access to a model posting state-of-the-art results on vulnerability detection, exploitation and patching benchmarks, while developers get Argon's $2-per-million-input-token introductory pricing unchanged. Google's staged Fairwind rollout and voluntary U.S. government testing keep public access gated until guardrail iteration is done.
Ask SkimNews
