MetaPatchAI home
AI

Google Unveils Gemini 4 Carbon, Its Strongest Model Yet, but Only Cyber Defenders Get It First

By MetaPatchAI Editorial Team · Published · Updated

Oct 30, 2026 — Google has a new top model, and most people can't use it yet. On Wednesday, Oct. 28, Google DeepMind announced Gemini 4 Carbon, which it calls "our strongest model" with state-of-the-art cybersecurity and long-horizon capabilities. Instead of a broad launch, Google is starting with a limited group of cyber defenders and expanding access gradually, a sign of how seriously frontier labs are now treating the security risks of their own models.

What Google is claiming

In the announcement, written by Koray Kavukcuoglu, SVP of Google DeepMind and Google's Chief AI Architect, Google makes several performance claims. All of these are Google's own reported figures:

  • Artificial Analysis Intelligence Index (v5.0): Ranked #1 with a score of 59, according to Google.
  • SEC-bench Pro (finding vulnerabilities): 88.2%
  • ExploitGym (exploitation): 54.8%
  • CWE-bench (patching): 71.0%
  • WebDev Arena: Up 89 Elo points to #4 (1768)

Google says the cybersecurity benchmarks were evaluated without guardrails. It also says Carbon delivers more than a 2x uplift over Gemini 4 Argon in generating secure code on internal benchmarks, and that Google's Chrome team reported Carbon performed more than 10x better than "another lab's large frontier model" at producing high-quality patches. Google did not name that model.

Built for long jobs

Carbon is pitched as an agentic workhorse. According to Google, internal teams have seen it "run for days" to push through dead ends on complex engineering tasks such as root-causing and patching subtle bugs across large codebases and low-level systems. Google also lists improvements in coding, scientific tasks, full-stack web development and multimodal understanding.

Safety and alignment claims

Google spends a big chunk of the post on safety, again citing its own testing:

  • Cheating on Terminal-Bench 4.0: Successful cheating reduced to 0.0%, with attempts down to 0.4%.
  • Misalignment alerts: High-severity misalignment alerts in live internal coding environments dropped by almost 80%.
  • Fewer unsupported claims: About 2x fewer than Argon in internal testing.
  • Scope discipline: Overreach down 26% and under-service down 29%.
  • Prompt injection: 0.0% attack success rate on Gray Swan's Indirect Prompt Injection Benchmark, versus 0.7% for Argon.

Who gets it, and what it costs

Google says it is taking a "deliberate approach" to rollout and is participating in the U.S. government's voluntary pre-release model testing process. Starting the week of the announcement, Carbon is going to a subset of cyber defenders, with access phased to more trusted partners through Google's Fairwind Program. Google says it will make Carbon available to developers, businesses and consumers "as soon as possible" after gathering feedback and refining guardrails.

Pricing, per Google:

  • Introductory: $2 per million input tokens and $10 per million output tokens (same as Argon's introductory price), with cached input 95% off.
  • After the intro period: $4 per million input tokens and $20 per million output tokens.

Carbon builds on Gemini 4 Argon, which Google says launched two weeks before Carbon.

Why it matters

Carbon's launch fits a broader pattern this week: frontier labs are increasingly framing new models around cybersecurity, both as a selling point and as a risk to manage. Google's staged rollout, starting with defenders rather than the general public, is a notable choice for a flagship model.

What's confirmed / not confirmed

Confirmed (Google, Oct. 28, 2026): - Gemini 4 Carbon was announced as Google's next frontier model. - Initial access is limited to a subset of cyber defenders, expanding via the Fairwind Program. - Pricing: $2/$10 per million tokens introductory, then $4/$20. - Google participates in the U.S. government's voluntary pre-release testing process.

Not confirmed: - A date for general availability to developers or consumers. Google says only "as soon as possible." - When the introductory pricing period ends. The post does not give an end date. - Independent verification of the benchmark and safety figures; all are self-reported by Google. - The identity of the "other lab's" model in the Chrome comparison.

Sources

Up: AI