Gemini 4 Carbon: what Google announced and who can use it

Gemini 4 Carbon is Google's frontier model announced on October 28, 2026, with an initial rollout to selected cyber defenders and trusted partners. Google describes improvements in software engineering, long-running tasks and cybersecurity. The announcement does not mean Carbon is already available to every Gemini user or developer.

As of October 29, Carbon is not selectable in Yix. Google's official announcement describes broader developer, business and consumer access as a later step. This guide explains the public information and what it means if you are choosing a model for creative work today.

Availability comes before a benchmark comparison

Google says it is beginning with a subset of cyber defenders, then phasing access to additional trusted partners through its Fairwind Program. It is gathering feedback and refining safeguards before broader distribution. The announcement does not give a general public release date.

If you want to use Carbon, start with the official announcement and the access information linked there. A model appearing in a ranking, video or news story does not establish that your account can run it. Check the actual model selection in the product you plan to use and its current access conditions.

This is also why Carbon should not be confused with Gemini 4 Argon. Google describes Carbon as building on Argon, but they are separate model releases. The Gemini API model directory is a useful place to check which models Google currently documents for developers. Access to Argon does not by itself grant access to Carbon.

What Google says changed

Google's announcement emphasizes sustained engineering work, coding and security tasks. It also describes improvements in web development and multimodal understanding. Those are the developer's reported capabilities, rather than results from a Yix evaluation.

The useful question for a team is narrower than “Is this the best model?” Define a task and an acceptance check. For a design implementation, that might mean translating a supplied mockup into a page, preserving responsive behavior and passing existing interaction checks. For a written brief, it might mean retaining all constraints across revisions without inventing product facts.

Model capability is only one part of those tasks. File access, code execution, browser controls, persistent state and approval rules come from the application around the model. A conversational interface does not automatically gain all of those features when it adopts a newer model.

Announced pricing is not a Yix offer

Google announced introductory pricing of $2 per million input tokens and $10 per million output tokens, with cached input priced at a 95% discount to the input rate. The same announcement's footnote says the rates after the introductory period will be $4 input and $20 output per million tokens. Check Google's current terms before making a budget; the announcement does not specify the introductory period's end date.

These are announced model rates. They are not a Yix subscription, credit quote or promise of access. A separate product may apply its own limits and pricing. In Yix, use the pricing page and the credit quote for the specific generation task you choose.

What this means for image and video creators

Carbon's announcement is primarily about reasoning and agentic work. It does not establish a new image-generation or video-generation mode in Yix. Understanding an image, writing a visual brief and rendering a finished image are different operations.

For example, suppose you need a product image with a clean background and a matching short video. You can first prepare the requirements: which product details must remain accurate, which background is allowed, what framing you want, and what the camera should do. Then choose the image or video model that supports those inputs and output settings.

A newer assistant may help formulate those instructions. The generation model still determines the actual rendering capabilities and limits. Before generating, inspect the selected model and parameters on the task card, rather than assuming the assistant's name identifies the renderer.

A practical brief you can prepare now

You do not need access to Carbon to make your next creative task easier to evaluate. Save a brief with a reference you have permission to use:

Use this product photograph as the reference. Preserve the product silhouette, cap color and label placement. Prepare a square studio-image prompt with soft side lighting and a neutral background. Then describe a five-second camera move for a video starting from that image. List any product detail that needs confirmation before generation.

This is an original planning example, not a Carbon output. Check each item in the resulting brief against the reference. If you later compare assistants, use the same input and acceptance criteria. Count substantive corrections, such as changed geometry or invented text, separately from preferences about writing style.

For an assistant already available in Yix, read the MiniMax M3.1 Flash guide. It explains how to choose that model for text and image discussion while keeping image and video generation separate. If your prompt is ready, you can go directly to the image generator or video generator.

What to check when wider access opens

Before moving a workflow to Carbon, verify the exact model shown in the product, supported inputs, account eligibility and current price. Run a small task with clear success criteria and inspect the output. A staged announcement is useful product news; it is not evidence that a particular application has completed an integration.

We will describe Carbon as available in Yix only after there is a usable integration. Until then, the official announcement is the source for rollout updates, and Yix's model menus show what you can actually select.