Particle.news

Google Releases Gemini 4 Carbon, a Frontier Model Focused on Cybersecurity and Long‑Horizon Tasks

The company will limit initial access to vetted cyber defenders to reduce misuse risk during a safety feedback period.

Overview

  • Google announced Gemini 4 Carbon on Wednesday, October 28, 2026, calling it a “frontier” update that extends Gemini 4 Argon with longer, multi‑step agent workflows that can run for days to root‑cause and fix complex software bugs.
  • Google says Carbon leads key benchmarks, reporting a #1 rank on the Artificial Analysis Intelligence Index and scores of 88.2% on SEC‑bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploit validation, and 71.0% on CWE‑bench for producing working patches.
  • The company describes specific safety and quality gains: internal tests show more than a twofold rise in secure code generation over Argon, large reductions in unsupported claims and misalignment alerts in live coding, and 0% success on an indirect prompt‑injection test.
  • Access will be phased: Google will start this week with a restricted rollout to selected cybersecurity defenders and Fairwind partners, delaying broad developer and consumer availability until it collects partner feedback and completes voluntary U.S. pre‑release testing.
  • Many claims come from Google‑run or partner evaluations, so independent verification of Carbon’s sustained agentic workflows and some benchmark results is limited; the model’s defender benefits could speed patching while the same capabilities raise dual‑use concerns about enabling misuse.