TL;DR Google announced Gemini 4 Carbon on October 28, 2026, calling it its strongest frontier model with state-of-the-art cybersecurity and long-horizon capabilities[1]. It ranks #1 on the Artificial Analysis Intelligence Index v5.0 at 59, and leads defensive-security benchmarks including SEC-bench Pro (88.2%)[1]. Like Argon before it, Carbon rolls out first to a subset of cyber defenders, not as a broad consumer launch[1].
Cybersecurity analyst working at a screen of code in a dark room, representing Gemini 4 Carbon's defensive security focus

The short answer

Gemini 4 Carbon is Google's second Gemini 4-generation model in two weeks, following Gemini 4 Argon[1]. Where Argon introduced a 1M-token output limit and defensive-cybersecurity skills, Carbon doubles down on that direction: it's optimized for sustained, long-horizon agentic work and for finding, validating, and patching software vulnerabilities[1].

Google is again taking a phased rollout — vetted cyber defenders first, then trusted partners, with developers, businesses, and consumers to follow "as soon as possible"[1]. So unless you're in that early cohort, Carbon is an announcement to watch rather than a model you can use today.

Pricing

Gemini 4 Carbon launches at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input price[1]. As with Argon, treat this as a preview-stage signal rather than a final price — broader availability is still pending guardrail iteration[1].

Benchmarks

  • Artificial Analysis Intelligence Index v5.0: ranks #1 overall at 59[1].
  • SEC-bench Pro: 88.2% for autonomously finding real-world vulnerabilities[1].
  • ExploitGym: 54.8% for validating whether a vulnerability is exploitable[1].
  • CWE-bench: 71.0% for turning findings into working code fixes[1].
  • WebDev Arena: #4 at 1768, a +89 Elo jump for full-stack web development[1].

Long-horizon coding and agents

Carbon is built for harder, longer engineering problems. Google says its internal teams have seen the model run for days, working through dead ends to root-cause and patch subtle bugs across large codebases and low-level systems[1]. On Terminal-Bench 4.0, Google reports Carbon reduced successful cheating to 0.0% (with attempts dropping to 0.4%) while still improving overall accuracy[1].

For most readers, though, the practical takeaway is simpler: Carbon is aimed at autonomous, multi-step software work — not quick chat answers.

Cybersecurity defense

The headline capability is defensive security across the full lifecycle — detection, validation, and remediation[1]. Google reports more than 2x uplift in generating secure code over Argon, and says the Chrome team saw Carbon perform more than 10x better than another lab's frontier model when generating high-quality patches[1]. Cloud-security firm Wiz also reports Carbon leading on web penetration-testing benchmarks used by its Scan for Good initiative[1].

On safety, Google highlights a 0.0% success rate against indirect prompt-injection attacks in its evaluation — resisting every attack attempt, versus Argon's already-low 0.7%[1].

Availability and rollout

Carbon is not broadly available yet. Google is engaged in the U.S. government's voluntary pre-release model-testing process and, starting the week of the announcement, is making the model available to a subset of cyber defenders, then phasing access to additional trusted partners through the Fairwind Program[1][3]. Developers, businesses, and consumers come later[1].

What it means for you

  • You can't sign up today. Unless you're in the Fairwind Program cohort, Carbon isn't accessible yet[1][3].
  • Security and platform teams should track it. The SEC-bench, ExploitGym, and CWE-bench results — plus Wiz's early use — point at vulnerability discovery and remediation as the primary use case[1].
  • Price is a preview signal only. At $2/$10 per million tokens, Carbon matches Argon's introductory pricing; final pricing is still to come[1].

Bottom line

Verdict: Gemini 4 Carbon is a frontier-model announcement with an even narrower launch than Argon — cyber defenders first, everyone else later. For now it's a signal of Google's aggressive pace on long-horizon reasoning and defensive security, not a product most readers can actually try.

Frequently asked questions

Can I use Gemini 4 Carbon right now?

Only if you're part of the initial cohort of vetted cyber defenders. A broader release to developers, businesses, and consumers is coming later[1].

How is Carbon different from Gemini 4 Argon?

Carbon builds on Argon with a stronger focus on sustained, long-horizon agentic work and defensive cybersecurity, and posts higher security benchmarks across SEC-bench Pro, ExploitGym, and CWE-bench[1].

How much will Gemini 4 Carbon cost?

The introductory API price matches Argon: $2 per million input tokens and $10 per million output tokens, with cached input at 95% off[1].

Sources

  1. Google Blog — "Gemini 4 Carbon: our next frontier model" (Oct 28, 2026) — blog.google
  2. Google DeepMind — "Gemini models" — deepmind.google
  3. Google DeepMind — "Fairwind Program" — deepmind.google
  4. Unite.AI — "Staged Rollout Sends Google's Gemini 4 Carbon To Cyber Defenders First" — unite.ai