Niko (@nikolayxyz) on X
Gemini 4 Carbon is Google's new frontier model, announced in a post by nikolayxyz that quotes Google's launch text. It ranks #1 on v5.0 of the Artificial Analysis Intelligence Index with 59 points, ahead of Opus 5.5 and GPT-6 Astra. Intro pricing is $2 input and $10 output per million tokens. Access starts for cyber defenders first, as Gemini 4 Argon's did. The captured text is cut off before the rollout details.
What it claims
Gemini 4 Carbon is Google's new frontier model. Google's launch text, quoted in a post by Niko (@nikolayxyz), says it ranks #1 on v5.0 of the Artificial Analysis Intelligence Index. Google also says it sets a new state of the art in cybersecurity and long-horizon capabilities, but the captured text gives no benchmark or score for either claim. Niko reports the index score as 59 points, ahead of Opus 5.5 and GPT-6 Astra, and gives introductory pricing of $2 per million input tokens and $10 per million output tokens.
How to use it
Niko says access starts "this week" for cyber defenders only. The captured text gives no API model name, endpoint, signup route or eligibility criteria for that group. It also does not show what "intro" pricing means: no end date or later price is stated.
For planning general access, Niko cites one precedent. Gemini 4 Argon launched about two weeks before Carbon, also to defenders first, and went public 13 days after that defenders-only launch. This is Niko's observation about Argon, not a date Google commits to for Carbon. If Carbon follows the same pattern, public access would come roughly two weeks after the defenders-only start. The post itself is undated in the captured text, so that window cannot be fixed to a calendar date from this page.
Limits
The source is a social post that summarizes and partly quotes Google. The index score, the competitor ranking and the pricing appear in Niko's words, not in the quoted Google text. Google's quoted text stops at "It's rolling out starting", so Google's own rollout schedule, regions and access terms are missing. The page gives no model size, context length, per-domain results, or the scores of Opus 5.5 and GPT-6 Astra on the index. Niko also says they have not been able to test every new model, and the post reports no hands-on testing of Carbon.
Organisations named here
Related in models


