Gemini 4 Carbon Tops AI Index, Yet Most Paying Customers Can't Buy It

On Wednesday, Google's Gemini 4 Carbon claimed the top spot on the Artificial Analysis Intelligence Index, an independently run ranking of AI systems. Even so, the model is unavailable to most paying customers.

Carbon posted 59, narrowly ahead of Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57. Alphabet (GOOGL) shares gained about 1.5% that day.

Why Is Gemini 4 Carbon Out of Reach for Most Buyers?

At first, Carbon goes to cyber defenders who pass screening under Google's Fairwind Program. Developers, businesses and consumers will be next, but Google has not set a date.

The phased release prioritizes safety, according to Google. Carbon's widest margin comes on hacking-related tests.

ExploitGym measures whether a model can demonstrate that a software flaw can be used in an attack; Carbon posted 54.8% there, while GPT-6 Astra recorded 42.4% and Opus 5.5 finished at 34.5%, per Google's benchmark table.

The non-cyber tests were run by Google with safety filters disabled. The company is also joining the US government's voluntary pre-release model testing.

Gemini 4 Argon, Google's previous top model, followed the same approach. It went live for paying developers on October 13, with Carbon arriving roughly two weeks later.

What Will Customers Pay for Gemini 4 Carbon?

Models such as Carbon are billed by Google on a per-token basis; tokens are the units of text the model processes and outputs.

Initial pricing is $2 per million input tokens and $10 per million output tokens. Later, those figures double to $4 and $20.

OpenAI, on the other hand, reduced GPT-6.1 Terra pricing to $1 and $5 on October 26. Anthropic, the company behind Claude, is getting ready for its planned stock listing.

Carbon's margin, though, is narrow. Most head-to-head numbers originate from Google's own table, and the public leaderboard at Artificial Analysis had yet to include Carbon as of Wednesday's check.

Some benchmarks still go to rivals. In FrontierSWE 2, a coding benchmark, Astra hit 65.5% and Opus 62.3%, with Carbon at 61.2%.

Google has not disclosed when access broadens or when the higher rates start. Meanwhile, all three labs have rolled out new models this month.