Imagera AI - AI content creation platform for generating images, cloning voices, and enhancing videos. Privacy Policy | Terms

AI & Technology

Gemini 4 Carbon: Google's Top Model, Not Yet Public

Illustration for the Gemini 4 Carbon language model announcement

Compare answers from several AI models in one Sandbox conversation.

AI Model Arena

TL;DR

Google announced Gemini 4 Carbon on October 28, 2026. It tops one independent intelligence index, but only a small group of cyber defenders can use it now, with no public API or app yet.

Key takeaways

  1. Google announced Gemini 4 Carbon on October 28, 2026.
  2. It tops one independent intelligence index, but only a small group of cyber defenders can use it now, with no public API or app yet.

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's latest frontier AI model. Google announced it on October 28, 2026, about two weeks after Gemini 4 Argon.

You can't try it yet. Google has not opened a public Gemini API or consumer app path for Carbon.

What is new in Gemini 4 Carbon?

Cybersecurity is the focus. Google reports that, evaluated without guardrails, Carbon scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench.

That "without guardrails" detail helps explain the limited release. Google says it needs more guardrail iteration before wider access.

Carbon also improves at building web apps. It jumps +89 Elo points to #4 (1768) on WebDev Arena, a ranking of full-stack web development quality.

Who can use Gemini 4 Carbon, and when?

Access is narrow. Starting the week of the announcement, Carbon is available only to a subset of cyber defenders. It will then reach trusted partners gradually through Google's Fairwind Program.

If you are a developer, a business, or a consumer, you don't have Carbon on any product surface yet. Google says that changes "as soon as possible" after more guardrail iteration. It has not given a date.

How much does Gemini 4 Carbon cost?

The introductory API price matches Argon at $2 per million input tokens and $10 per million output tokens. Cached input tokens get 95% off the input price.

The price will go up later. After the introductory period, Carbon costs $4 per million input tokens and $20 per million output tokens. Google has not stated an end date for the introductory rates.

If you plan to build on Carbon, budget for the higher rates. Those prices apply only to the limited set of approved users for now.

How does Gemini 4 Carbon compare with other models?

On independent measures, Carbon sits at or near the top. It ranks #1 on the Artificial Analysis Intelligence Index v5.0 with a score of 59.

For web development, its #4 position on WebDev Arena shows it is strong but not leading there. These are results Google reported, and you can't yet check them yourself with your own workloads.

What can you use today?

Gemini 4 Carbon is not publicly available yet. Imagera adds new models once they have a public API; until then, these cover the same jobs:

  • LLM Arena: send one prompt to several current language models and compare the answers side by side.
  • Imagera Code: a coding agent that works in your project with the model you pick.

Frequently Asked Questions

How do organizations get into the Fairwind Program?
You apply through a Google Cloud account team. Priority goes to governments, critical infrastructure, and core technology platforms.
What can Carbon take as input and return?
Artificial Analysis lists a 1M-token context window. It accepts text, image, speech, and video input and returns text.
Has Carbon been through government safety testing?
Google is in the U.S. Government's voluntary pre-release model testing process.

Imagera AI Team

The editorial byline of Imagera

Articles are published under one team byline. Drafts may be written with AI assistance; a person on the Imagera team checks facts, links and images and publishes each one.

Areas of Expertise:

AI Image GenerationAI Video CreationAI Voice & AudioLoRA Model Training

Cite this page: https://imagera.ai/blog/gemini-4-carbon. Name Imagera AI as the source when you quote it.

Create without limits

Compare answers from several AI models in one Sandbox conversation.

Open AI Model Arena →