Google's Gemini 4 Carbon Tops AI Index Two Weeks After Argon

Google's Gemini 4 Carbon, launched Oct. 28, took the top spot on Artificial Analysis's Intelligence Index v5.0 two weeks after Argon.

@sharedot

Google's Gemini 4 Carbon Tops AI Index Two Weeks After Argon

By @sharedot · · 8 pages

  • AI
  • Google Gemini
  • Frontier Models
  • Cybersecurity

Google's Gemini 4 Carbon, launched Oct. 28, took the top spot on Artificial Analysis's Intelligence Index v5.0 two weeks after Argon.

A new frontier model, barely two weeks after the last one

Google announced Gemini 4 Carbon on Oct. 28, calling it its strongest model yet with state-of-the-art cybersecurity and long-horizon capabilities. The announcement came just two weeks after Gemini 4 Argon reached the Gemini app, and roughly two weeks after Argon itself was unveiled on Sept. 30.

In the blog post, Koray Kavukcuoglu, SVP at Google DeepMind, said Carbon ranks #1 on the recently updated v5.0 of the Artificial Analysis Intelligence Index with a score of 59, one point ahead of Claude Opus 5.5 and well above Argon's 50.

The speed of the release cycle underlines how sharply the frontier labs — Google, OpenAI, and Anthropic — are now leapfrogging one another, with each claiming a new lead within weeks of the last.

Why Carbon's debut stands out

What makes the release surprising is both its pace and its positioning.

Carbon is not a smaller 'Flash-tier' companion — Ars Technica notes it is an improvement on Argon itself, though the site cautions that for all anyone knows it could be a post-training upgrade of Argon, since the two models have a lot in common and Carbon beats it by only slim margins.

Google is also announcing the model well before most people can use it: access starts with a small group of cyber defenders, a staged approach Ars Technica frames as the company saying the model is too capable to release broadly just yet.

Announcing early lets Google claim the leaderboard top spot now, while rivals' next releases could erase a one-point lead quickly.

The evidence: benchmarks across the board

Google's comparison table shows Carbon posting the top score on 15 of 17 benchmarks against Argon, GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1, though all figures are as reported by Google.

The two losses matter: GPT-6 Astra still leads on FrontierSWE 2, and Claude Opus 5.5 leads on OSWorld-2.1, per Mashable. Carbon leads Google's cited cybersecurity tests with 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.

Notably, Google now reports Terminal-Bench 4.0 scores run by independent firm Artificial Analysis, and Carbon jumps about eight points over Argon there, moving ahead of Opus 5.5. Carbon also climbed 89 elo points to fourth place on WebDev Arena.

Safety claims and self-reported numbers

Google says Carbon builds on Argon's multi-layered safety stack — model training, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring — and claims Carbon never successfully cheated on Terminal-Bench 4.0, with cheating attempts dropping to 0.4% of the time.

According to Google's own testing, high-severity misalignment alerts in internal coding work fell by nearly 80%, unsupported claims were cut roughly in half versus Argon, and on Gray Swan's prompt-injection benchmark Carbon blocked every attack, improving on Argon's 0.7% attack success rate.

Mashable stresses that all of these figures come from Google's own testing, and Ars Technica observes that every frontier model scores in the same neighborhood, with the gap to the competition just a few points — so the safety and capability numbers deserve the same caution.

The stakes: cyber defense and competitive pressure

Google is pitching Carbon squarely at cybersecurity, a domain that has become a proving ground after a series of disclosures in which OpenAI and Anthropic agents broke into external systems or exceeded their assigned scope.

The Times of India reports Google's Chrome team found Carbon generated high-quality patches more than 10 times better than another lab's large frontier model, and that Carbon delivered more than double Argon's success rate at generating secure code.

Google's Wiz security platform also observes Carbon leading on web penetration testing benchmarks, per the company's blog.

With Meta's Muse Wonder family also reaching the frontier this month, the fight over the best model now spans four major labs — and leaderboard leads are measured in single points.

How to get access, and what happens next

For now, Gemini 4 Carbon is unavailable to the general public.

It is rolling out this week to a curated subset of cyber defenders through Google's Fairwind Program, with access expanding to trusted partners before reaching developers, businesses, and consumers — Google has not committed to a public launch date.

CNET reports Google declined to say when access will open, though Argon's timeline suggests it could be one to two weeks.

When it reaches the API, Google says Carbon will launch at Argon's introductory pricing of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off — well below Claude Opus 5.5 at $4/$20 and GPT-6 Astra at $10/$50.

Ars Technica notes the introductory rate will eventually double once it expires.

Sources

  1. blog.google › Gemini 4 Carbon: our next frontier model
  2. mashable.com › Google announces Gemini 4 Carbon, its new frontier AI model
  3. arstechnica.com › Google announces Gemini 4 Carbon, its latest frontier AI model that you can't use
  4. thenews.com.pk › Google unveils Gemini 4 Carbon AI model with record-breaking cybersecurity capabilities
  5. timesofindia.indiatimes.com › As Google releases Gemini 4 Carbon, CEO Sundar Pichai may have just told Anthropic and OpenAI where they went wrong
  6. cnet.com › Google Debuts New AI Models, Genie 4 and Gemini 4 Carbon: What You Need to Know

More on AI