Google's new AI edges out Claude by one point, but you can't use it yet

Google's newest AI model, Gemini 4 Carbon, scored 59 on a respected independent AI test. Anthropic's Claude Opus 5.5 scored 58. The catch is access. Carbon is not open to the public, and Google has so far offered it only to selected cyber defenders and trusted partners.

What the scores say

The test is the Artificial Analysis Intelligence Index, version 5. It averages nine tests covering knowledge work, coding, science and general ability. Carbon posted the top score in four of the nine. Opus 5.5 kept the lead on both coding tests. On coding agents, which are AI tools that write and run code on their own, Opus 5.5 leads 62 to 59. OpenAI's GPT-6 Astra sits at 57.

Why one point is thin

The two models were not run at the same effort setting. Carbon was measured at high effort and Opus 5.5 at max. The test suite is also new, and about 60% of it uses private tests or withheld answers, so it is hard to compare with older charts. Google's own comparison chart shows no uncertainty estimates. A one-point gap on a first run is inside any reasonable noise band. Think of it as a tie with a ranking attached.

What you can actually use

Google reportedly announced Carbon on October 28, 2026, and access runs through its Fairwind Program, with broader access still pending. The Gemini most people can use is Gemini 4 Argon, reportedly generally available since October 13. Argon scored 50 on the same index. Opus 5.5 is available on the Claude API, Amazon Web Services, Google Cloud and Microsoft Azure.

On cost, Carbon runs $7.76 per index task against $10.04 for Opus 5.5. Argon costs $7.68. Opus 5.5 is priced at $4 and $20 per million tokens, the chunks of text an AI reads and writes.

What to do

• If you build or buy software, test Opus 5.5 and Argon on your own prompts.
• If you do security research and hold a Fairwind seat, Carbon is the one to try.
• If you only chat in an app, the plan you already pay for decides most of it.