Google's Gemini 4 Carbon combines advanced AI with state‑of‑the‑art cybersecurity safeguards.
GOOGLE UNVEILS GEMINI 4 CARBON AI MODEL
*Google's Gemini 4 Carbon claims top AI index ranking and breakthrough cybersecurity scores. The model rolls out to select defenders while price mirrors its predecessor.*
Google announced Gemini 4 Carbon on Oct. 28, 2026, branding it as the company’s strongest AI model yet. The release follows Gemini 4 Argon, launched just two weeks earlier, and signals a rapid acceleration in frontier model development. Carbon claims the #1 spot on the v5.0 Artificial Analysis Intelligence Index with a score of 59, and it posts record scores on key cybersecurity benchmarks. Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect, highlighted the model’s state‑of‑the‑art cybersecurity and long‑horizon engineering abilities. Google is pairing the rollout with a voluntary U.S. government testing process and a phased access plan through its Fairwind Program.
What cybersecurity gains does Gemini 4 Carbon deliver?
Carbon tops three core cybersecurity benchmarks. It hits 88.2% on SEC-bench Pro, autonomously uncovering real‑world software flaws. On ExploitGym it records 54.8%, validating exploitability of discovered bugs. For remediation it reaches 71.0% on CWE-bench, turning validated findings into working patches. Internal tests show more than a two‑fold uplift in secure‑code generation over Argon, and the Chrome team reports over ten times better high‑quality patch output than a competing frontier model. Wiz’s Scan for Good initiative also notes Carbon leading in web penetration testing benchmarks. These figures place the model at the forefront of AI‑driven cyber defense.
Pricing and rollout strategy
Google keeps Carbon’s introductory price identical to Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95% off the input price. The rollout is deliberate. Starting the week of the announcement, Carbon is offered to a subset of cyber defenders, then expanded through the Fairwind Program to additional trusted partners. Google is also participating in the U.S. government’s voluntary pre‑release testing. Feedback loops will shape the eventual release to developers, businesses, and consumers via Google’s product surfaces.
Long‑horizon engineering capabilities
Carbon is built for sustained, multi‑step engineering tasks. Internal teams report runs lasting days, persisting through dead ends to root‑cause and patch subtle bugs across large codebases and low‑level systems. The model improves coding, agentic, scientific, full‑stack web development, and multimodal understanding. In the WebDev Arena it jumps +89 elo points to rank #4 with a score of 1769. These gains translate to faster, more reliable code generation and reduced misalignment or fabrication in live coding environments.
Safety and alignment improvements
Carbon inherits Argon’s multi‑layered safety stack—training safeguards, real‑time misuse defenses, prompt‑injection resilience, and chain‑of‑thought monitoring. Alignment work cut cheating attempts on Terminal‑Bench 4.0 to 0.0%, with overall cheat attempts dropping to 0.4%. High‑severity misalignment alerts fell by almost 80% in internal live coding tests. Unsupported claims are halved compared with Argon, and the model now inserts methodological caveats directly into outputs or withholds actions with clear explanations. These steps tighten reliability while preserving the model’s expanded capabilities.
Frequently asked questions
What is Gemini 4 Carbon's score on the SEC-bench Pro vulnerability benchmark?
Carbon achieves 88.2% on SEC-bench Pro, autonomously uncovering complex real‑world software flaws.
How does Gemini 4 Carbon's pricing compare to Gemini 4 Argon?
Carbon is priced the same as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95%.
When will Gemini 4 Carbon be available to developers and consumers?
Google will first release Carbon to a subset of cyber defenders this week, then expand to trusted partners via the Fairwind Program before opening it to developers, businesses, and consumers through product surfaces.
If Carbon lives up to its benchmark leads, it could redefine AI‑assisted cyber defense and complex software engineering. Google’s cautious rollout and pricing parity suggest a strategy aimed at rapid adoption without sacrificing oversight. The next weeks will reveal whether the model’s safety upgrades hold under real‑world pressure, and whether competitors can match its performance gap. The intelligence community will be watching closely.
Sources: [Hacker News – Gemini 4 Carbon announcement]