AI News2026 / 10 / 306 min readMembers only

Google's Gemini 4 Carbon Ships to Cyber Defenders First; Everyone Else Waits

Announced October 28 by Chief AI Architect Koray Kavukcuoglu as building on Gemini 4 Argon, Carbon tops the v5.0 Artificial Analysis Intelligence Index at 59 and climbs 89 elo points to fourth on WebDev Arena. Its security benchmarks were run explicitly without guardrails: 88.2% on SEC-bench Pro, 54.8% on ExploitGym, 71.0% on CWE-bench. Cyber defenders get access first through the Fairwind Program; developers and consumers wait with no date given. Watch pricing double to 4 and 20 dollars per million tokens after the intro period, and treat every figure as vendor-reported.

XXingchen Desk· 2026 / 10 / 30 · 6 min read · Members only

At a glance

  • 1Ranks #1 on Artificial Analysis Intelligence Index v5.0 at 59; jumps +89 elo to #4 (1768) on WebDev Arena
  • 288.2% on SEC-bench Pro, 54.8% on ExploitGym, 71.0% on CWE-bench — all evaluated without guardrails
  • 3Introductory pricing $2/$10 per million input/output tokens, rising to $4/$20 after the intro period

What shipped

On October 28, 2026, Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, announced Gemini 4 Carbon — described as "our strongest model with state-of-the-art cybersecurity and long horizon capabilities." It builds on Gemini 4 Argon, released two weeks earlier.

The "long horizon" framing is concrete rather than marketing. Internally, Google says teams have watched Carbon run for days, push through dead ends, and handle tasks like root-causing and patching subtle bugs across large codebases and low-level systems. That is a description of an agent that works, not a chatbot that answers.

The numbers

On aggregate rankings, Carbon takes #1 on the recently updated v5.0 of the Artificial Analysis Intelligence Index at 59, and climbs +89 elo points to #4 (1768) on WebDev Arena.

The security benchmarks were run without guardrails, which Google states explicitly:

  • SEC-bench Pro (finding real vulnerabilities buried in real-world software): 88.2%
  • ExploitGym (validating risk through exploitation): 54.8%

Members only · the rest is locked

Become a member to unlock full articles and the complete podcast feed

See plans

Full-text Markdown

Copy or download in one click, then paste into notes or feed it to an AI

Download .md