Kevin's avatar
Kevin
Part time degen

Google's new frontier model leads the board. You cannot use it yet.

Oct 29, 2026. About Gemini 4 Carbon, cybersecurity, and staged releases.

On October 28, 2026, Google announced Gemini 4 Carbon, its next frontier model after Gemini 4 Argon (Google blog). Carbon sits at the top of a major independent leaderboard. It also posts strong scores on vulnerability finding, exploitation, and patching. Google is not giving developers, businesses, or consumers access this week. The first users are a subset of cyber defenders, through Google's Fairwind Program, while Google finishes U.S. government voluntary pre-release testing and more guardrail work.

That gap between the scoreboard and the API is the story. A model that is good at finding and exploiting software flaws is also a model you do not hand to everyone on day one.

What the independent scoreboard says

Artificial Analysis runs the Intelligence Index v5.0, a composite of nine evaluations across knowledge work, coding, science, and general tasks. On October 28 it put Gemini 4 Carbon (High) first at 59, one point above Claude Opus 5.5 (Max) at 58 (Artificial Analysis). Carbon led four of the nine Index evaluations. On Artificial Analysis's separate Coding Agent Index, Carbon ranked second, behind Claude Opus 5.5.

Those numbers come from an outside lab, not from Google's marketing deck. They are the cleanest public evidence that Carbon moved the frontier this week.

What Google claims on cybersecurity

Google's post centers Carbon on the cyber defense lifecycle: find a bug, check that it is exploitable, then patch it. The numbers Google published, evaluated without guardrails, are:

Google also reports internal results I could not verify independently. It says Carbon more than doubles Argon's success rate at generating secure code. It says the Chrome team found Carbon more than 10 times better than another lab's large frontier model at high-quality patches. Wiz, Google's cloud security platform, says Carbon leads on its web penetration testing benchmarks. Treat those as vendor and partner claims until someone outside Google repeats them.

Alignment numbers Google wants you to notice

Alongside capability, Google published several safety metrics from its own tests.

Again, these are Google's measurements. Ars Technica notes that Carbon beats Argon by slim margins on Google's charts and that every frontier model sits in a tight band (Ars Technica). The independent Index lead is clearer than the size of the cyber gap.

Price is set. Availability is not

Google listed the same introductory API price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input at 95% off. After the introductory period, the price becomes $4 and $20. Google did not publish an end date for the discount. Ars asked and Google declined to give one.

Access is gated. Google says it is in the U.S. government's voluntary pre-release testing process. This week's cohort is cyber defenders. More trusted partners follow through Fairwind. Developers, businesses, and consumers come later, "as soon as possible," with no date. Argon launched about two weeks before Carbon, so Google's recent pattern is short cycles between named frontier drops and longer waits before open API access.

My view

Carbon matters for two reasons that pull in opposite directions.

First, an independent lab put it on top of Index v5.0 the day Google announced it. That is a real capability move, not only a press release. If you care about coding agents and long-horizon engineering work, Carbon is the model to watch when the API opens.

Second, Google's own cyber numbers, and the Fairwind-first rollout, say the risk case is no longer theoretical. A model that scores high on ExploitGym is useful to defenders and dangerous in the wrong hands. Staged access is the right call. The open question is whether Fairwind feedback and the U.S. voluntary process produce a clear public safety report before general availability, or whether Carbon just appears in the API one day with little new documentation.

For builders, do not plan a product on Carbon this week. Watch Fairwind partner reports and the day Artificial Analysis or another lab can run Carbon without partner access. For markets, Argon's two-week gap to Carbon already showed Google's release cadence. Carbon shows Google is willing to hold a #1 model back when the cyber scores get sharp. Nothing here is financial advice.

Sources