Author｜Sangbi District Shi Editorial Team
Release｜2026/10/29
Update｜2026/11/11

Gemini 4 Carbon is the Gemini series of AI models announced by Google on October 28, 2026. It focuses on program engineering, information security defense, and AI agent tasks that require continuous execution of multiple steps.

Gemini 4 Carbon was officially opened on the Gemini API on November 10, 2026, with the model code gemini-4-carbon. The API opening and the consumer plan of Gemini App are different entrances, so it cannot be inferred that all Gemini accounts can directly use Carbon. This article organizes its positioning, API price and usage information; general chat, free functions and Taiwan subscription methods, please refer to Gemini User Guide。

What is Gemini 4 Carbon? Let’s first look at the differences between models and products

Gemini 4 Carbon is the name of the model; Gemini web version and App are the product entrances using AI. If you understand this difference, you will not directly interpret "Google releases a new model" as "its own chat interface has been upgraded."

When using AI services, confirmation can be divided into three levels:

Is the problem model to be confirmed actually using Carbon, Argon, or other models? Product entry from Gemini App, development tools or API? Account benefits: Do my plans, regions, and accounts have corresponding qualifications and quotas?

The same product portal may provide multiple models; the functions of subscription plans with the same name in different products should also be checked separately. Therefore, if you want to use Carbon, the most direct basis for judgment is the model shown in the official opening announcement and actual accounts.

What features does Gemini 4 Carbon have?

Program engineering: from answering questions to completing the workflow

Google places important improvements in Carbon on complex program engineering and proxy tasks.

Officials stated that internal teams had observed Carbon running continuously for several days, tracing and patching issues in the large code base and underlying systems. This is an internal usage observation published by Google and does not mean that general accounts will be able to perform continuously for the same period of time.

Taking "fixing a website error" as an example, the complete work may include reading the problem, finding the relevant program, proposing modifications, performing checks and explaining the results. This requires more task connection and result verification than generating a piece of code at a time. This is an example to illustrate the agent workflow and is not an actual test of Carbon by this site.

Information security defense: focus on vulnerabilities and patching

Officials emphasize Carbon's capabilities in vulnerability detection, verification and patching. For the development team, the focus of the evaluation can include: whether the real problem can be found, explaining the cause of the problem, and proposing changes that pass the inspection.

This type of capability is also related to software security issues for cryptocurrency applications, but it cannot be directly presumed that the model has completesmart contractAuditing capabilities; general software security testing and professional auditing of specific chains and contract languages ​​need to be verified separately.

Long-term tasks: The focus is on continuous advancement and delivery

Long-term tasks refer to work that requires multiple rounds of processing, tool operation, and repeated checking. It does not mean simply writing long answers. During actual evaluation, you can observe whether the model can be corrected after encountering errors, whether important conditions are retained, and whether it delivers checkable results in the end.

What is the difference between Gemini 4 Carbon and Argon?

Both are Gemini 4 series models. The difference that currently has the most impact on general readers is the published competency direction and the open stage.

Comparison project Gemini 4 CarbonGemini 4 Argon officially announced that it focuses on strengthening program engineering, information security and continuous agent tasks, complex reasoning, program engineering and cross-domain work. The status as of 2026/11/11 Gemini API has been officially opened; Gemini App has also confirmed public products and API entrances based on product announcements and account qualifications. Gemini App has not yet announced that it will be fully open to general accounts. 10/13 The official announcement has listed Pro and Ultra subscriber entrance APIs. The specifications confirm that there is an official model file. The code is gemini-4-carbon, which can be found on the public model file and rate page.

Google announced the public rollout of Argon on October 13. Carbon will be released on October 28th, so don’t move Argon’s model code, context caps, feature support, or subscriptions directly to Carbon.

For those who want to start using it at this stage, it is more useful to first confirm which model is available than just comparing the old and new names.

Is the Gemini 4 Carbon available now? How to check the opening hours in Taiwan?

Google has announced that Gemini 4 Carbon is officially open in the Gemini API on November 10, 2026. Developers can use gemini-4-carbon according to the official model file. This announcement is for the API and does not mean that Gemini App has opened Carbon to all regions, plans and general consumers; Taiwanese users should still confirm the entrance used, account qualifications and available models respectively.

Taiwanese readers can confirm in the following order:

  1. Check to see if Google announces any other product opening news about Carbon.
  2. Confirm whether the announcement is talking about the Gemini App, API, or other development tools.
  3. Check whether your account can choose Carbon, as well as the applicable plans and quotas.
  4. Before using the API, confirm the model code, billing and function files.

If you just want to get started with Google AI right now, you don’t have to wait for Carbon; you can start by following Gemini usage tutorialComplete login and first task. Just because an existing plan is available does not mean you have achieved Carbon qualification.

How much does Gemini 4 Carbon API cost?

The Carbon API price announced by Google is as follows, the currency is US dollars, and the pricing unit is per 1 million tokens:

The initial price of the project launch (until 2026/11/30) After the initial discount ends (from 2026/12/01) input US$2US$4 and output US$10US$20

The cache entry price is US$0.10 per 1 million tokens until November 30, 2026, and US$0.20 from December 1. Carbon currently only supports implicit caching, not explicit caching, and does not offer a free tier for the Gemini API. The above is standard billing. Fees for other processing modes and search tools must be confirmed separately on the official price page.

How to estimate an API cost?

First separate the input and output usage, and then calculate based on their respective unit prices.

Assuming that the initial price is used, a job consumes 1 million input tokens and 100,000 output tokens, does not use cache, and does not include other costs:

Input US$2 + Output US$1 = US$3.

If the subsequent price in the table is used for the same usage, it will be US$4 + US$2 = US$6. This is an example of a conversion based on specified conditions and is not an actual test or a fixed quote for each job.

Token is not a fixed word count either: content, language and treatment will affect usage. Agent jobs may also repeatedly read data and call tools, so when comparing costs, look at the total cost to complete the entire job.

Does a Gemini subscription include the Carbon API?

Individual subscription rights and API billing should be confirmed separately. It cannot be inferred from the name of Google AI Pro or Ultra that the API usage of Carbon is included. Developers can view additional Gemini API official price pageand subsequent Carbon files.

What do you think of the official review? Why can't we just look at the score?

Model reviews need to be read together with test versions, tools, running times, and success determination methods.

Google's Carbon evaluation method document states that some scores are calculated by Google, and some refer to external rankings; other model information may also come from results published by the vendors themselves. The execution conditions of different test projects are not all the same.

Taking SEC-bench Pro as an example, the document explains the JavaScript engine question group used and the success determination; ExploitGym involves a specific tool environment and task time limit. They reflect specific test abilities but do not directly answer the experience of Traditional Chinese writing, daily chatting, or all projects.

When selecting a model, you can use small tasks that you are familiar with and record the correctness, completion time, cost, and areas that require manual modification. Only by comparing under the same tasks and conditions can it be closer to your actual needs.

Restrictions and precautions for using Gemini 4 Carbon

The current article is mainly based on official information compilation, and it is not an actual test report using Carbon on this site. After subsequently obtaining access rights, you should still confirm the available tools, data processing terms and work results separately.

If you want AI to process programs or documents, you can first use a test environment and non-sensitive data. The positioning of information security capabilities cannot be regarded as a promise that the program will have no loopholes, the contract will be safe, or the answer will be correct.

Gemini 4 Carbon Latest News

The significance of date events to users 2026/10/28 Google announced Carbon and phased opening arrangements. Full product opening still needs to wait for follow-up announcements. 2026/11/10 Gemini API is officially open. The model code is gemini-4-carbon; the standard input/output preferential price is US$2/US$10 per million tokens until November 30, and will be adjusted to US$4/US$20 from December 1.

For related events, please refer to this site Google releases Gemini 4 Carbon news. The follow-up updates of this area will focus on the official opening, functional specifications and price changes.

Gemini 4 Carbon FAQ

Is Gemini 4 Carbon a new app?

Carbon is the model name. Where to use it, you need to check the opening arrangement of the corresponding product.

Can Gemini 4 Carbon be used in Taiwan?

Gemini 4 Carbon is officially open on the Gemini API, but the API has different qualifications than the Gemini App. Taiwanese users should confirm the regions, accounts and payment conditions supported by the service they are using. The opening of the API cannot be directly regarded as having obtained Carbon for a general Gemini App account.

Is Gemini 4 Carbon free to use?

The Carbon free tier is not currently available on the Gemini API official pricing page. The free features of Gemini App and API billing are different arrangements and cannot be confused.

When will Gemini 4 Carbon be fully available?

Gemini API has been officially opened on November 10, 2026. Gemini App consumer eligibility still needs to be confirmed based on product announcements and actual account display.

Which one should I choose, Gemini 4 Carbon or Argon?

First confirm the models you can use, and then compare the quality, time-consuming and cost with the same task. Don’t just decide by name or a single ranking list.

Is Carbon's API price a monthly fee?

no. The price list in this article is calculated based on token usage, which is different from the monthly subscription fee for individuals.

Can I use Argon's model code to call Carbon?

Can't. The official model code of Carbon is gemini-4-carbon, which should be set according to the corresponding file; code using Argon will not automatically switch to Carbon.

Can Gemini 4 Carbon audit smart contracts?

Just relying on general information security capabilities cannot be deemed sufficient to complete professional smart contract audits. Additional confirmation of contract language, tools, testing and manual verification is required.

official sources

*This article was compiled by the Sangbi District Editorial Team based on official information and public information. The content is for information and educational purposes only and does not constitute investment advice.

Source