Google's #1 Model Has a Guest List. You're Not On It.
Google's Gemini 4 Carbon beat Claude and GPT-6 on the Artificial Analysis index two weeks after Argon shipped, but it's locked behind a government pre-release program with no date for paying customers.

Google's last "best model in the world" lasted fourteen days before Google itself made it obsolete. The one that replaced it is still locked in a vault.
On Wednesday, Oct. 28, Google announced Gemini 4 Carbon, its newest frontier model. (cite index="2-5">Carbon scores 59 on Artificial Analysis' index, one point above Claude Opus 5.5, ending months in which Google's best model trailed its rivals. (cite index="2-8">Carbon arrives 15 days after Gemini 4 Argon became broadly available, and it improves on Argon on every benchmark Google published. Two frontier models, fourteen days apart. Google's own engineers are the only ones who can verify both on paper.
| Index Score (points) | |
|---|---|
| Gemini 4 Carbon | 59 |
| Claude Opus 5.5 | 58 |
| GPT-6 Astra | 57 |
| Gemini 4 Argon | 50 |
Here's the catch. (cite index="2-11,2-12">Google is releasing Carbon to a subset of cyber defenders this week and will widen access gradually through its Fairwind Program, which has more than 650 partners, including government cyber authorities, critical infrastructure operators in healthcare, telecommunications and energy, and core technology platforms. (cite index="2-14">Fairwind members undergo background checks, must use phishing-resistant multi-factor authentication, may only give access to internal security, incident-response or penetration-testing teams, and may not share or sell model access. Google is also (cite index="3-19">"actively engaged in the U.S. government's voluntary process for pre-release model testing."
Translation: we beat the field, and the field doesn't get to find out firsthand.
The Benchmark Is Real, the Product Isn't
(cite index="5-15,5-16">Google's security numbers are vendor-reported: 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them, and 71.0% on Collinear's CWE-bench for patching. But (cite index="5-17">Google ran those external benchmarks without guardrails, which means the model your security team gets may score differently.
The weakest number is also the one that matters most. (cite index="4-23,4-24">At 54.8% on ExploitGym, Carbon is the weakest of the three security benchmarks by a wide margin. A model that's mediocre at proving an exploit works still needs human verification.
A leaderboard screenshot is not a production deployment, and Google knows it. (cite index="2-10">Its previous turns at the top lasted only weeks before OpenAI and Anthropic caught up. Carbon's benchmark crown has an expiration date baked in, and it spent its entire shelf life so far behind a government intake form.

| Input | Output |
|---|---|
| Introductory | |
| $2 | $10 |
| Standard (after intro period) | |
| $4 | $20 |
You Cannot Buy What Beats the Competition
(cite index="8-7">Carbon launches at Argon's introductory price of $2 per million input tokens and $10 per million output tokens, which rises to $4 and $20 when the introductory period ends. You get cheap tokens today for a model you cannot run, and a doubled bill later for whichever version Google finally ships to the API.
(cite index="8-2">Argon reached general availability 13 days after its Fairwind launch. The pessimistic case has a name: Anthropic's Mythos. (cite index="5-35">Argon went public in 13 days, and Mythos stayed gated. Nobody has told you which path Carbon takes.
If your procurement plan assumes a two-week wait, you're guessing. The embargo isn't a safety measure bolted onto the business. For Google's commercial pipeline, the embargo is the business.
What to watch: whether Carbon reaches the public API inside Argon's 13-day window or settles into Mythos-style limbo; whether Anthropic or OpenAI cut prices while Google's commercial shelf sits empty; and whether the next Gemini model displaces Carbon on the leaderboard before a single enterprise customer gets to run it.
- Gemini 4 Carbon AI Model Puts Google Back on Top, But Only Cyber Defenders Can Use It For Now
- Google's Gemini 4 Carbon AI model goes to cybersecurity defenders first
- Gemini 4 Carbon Tops Artificial Analysis, but Google Limits Access
- Google Launches Gemini 4 Carbon for Vetted Cyber Defenders First
- Gemini 4 Carbon: Benchmarks, Price and Release Date
- Google on X: "Introducing Gemini 4 Carbon, our latest frontier model. Carbon delivers improved performance across domains — ranking #1 on v5.0 of the @ArtificialAnlys Intelligence Index and sets a new state-of-the-art in cybersecurity and long-horizon capabilities. It’s rolling out starting this… / X
- Google's Gemini 4 Carbon pushes cyber defense first
- Gemini 4 Carbon rolls out first to cyber defenders
- Gemini 4 Carbon: Index 59, $2/$10 Price, Who Gets It
- Google introduces Gemini 4 Carbon, reserving first access for cyber defenders
- Gemini 4 Argon: Google is back as one of the top three labs in intelligence achieved
- Gemini 4 Argon: Benchmarks, Price and Who Gets It
- Google makes Gemini 4 Argon broadly available with limited-time low API pricing of $2/$10 per 1M token
- Gemini 4 Argon: What's New, Who Can Use It, Pricing, and Benchmarks
- Gemini 4 Argon Pricing: Cheap Per Token, Not Per Task
- Gemini 4 Argon Pricing: Is It Actually Cheaper Than Rival AI Models?
- Gemini 4 Argon: Pricing, Benchmarks, and Cyber Defense Rollout Explained - DevX
- Gemini 4 Argon: Price, Plans and How to Get It (2026)
- Artificial Analysis on X: "Google’s new Gemini 4 Argon equals GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices. Google is now back to being one of the top three labs in intelligence achieved Gemini 4 Argon is @GoogleDeepMind’s first propr… / X
- How Mythos-class AI is changing cyber security risk
- Anthropic will give defenders what its strongest model finds, but not the model itself
- How Anthropic's Mythos AI Model Could Change Cybersecurity
- What Is Mythos AI? Complete Technical Guide
- Too Dangerous to Deploy: Anthropic's Mythos and What Comes Next
- What Anthropic's Mythos AI Model Actually Means for Defenders: And Why the Window Is Closing
- Claude Mythos: Anthropic's Gated AI Model 2026
- Large Language Models and Attention-Based AI for Hardware Design and Security: Progress, Challenges, and Opportunities
- Artificial Intelligence as Game Changer in Cybersecurity: What We Learned in 2025-2026, and how this is relevant for Africa