Gemini 4 Carbon vs Argon: Which to Use, Plan Access and API Pricing
Updated November 11, 2026
On November 10, 2026, Google announced Gemini 4 Carbon for developers, businesses and eligible individual users. Carbon is a new high-capability model in the Gemini 4 family, aimed at difficult coding, long-running AI agents, scientific research and demanding knowledge work.
How does it differ from Gemini 4 Argon? Do Google AI Pro subscribers need Ultra to try it? Can developers use the API directly, and how much more might it cost?
Google positions Carbon as the more capable model, but that does not make it the right replacement for Argon on every task. This guide draws on Google's November 10 announcement, Gemini Help and developer pricing documentation to explain the differences, access rules and costs.
1. What is Gemini 4 Carbon, and how does it differ from Argon?
Carbon is the Gemini 4 family's flagship reasoning model for difficult tasks. Google's positioning is:
- Gemini 4 Argon: most software engineering, knowledge work and routine complex tasks.
- Gemini 4 Carbon: problems that need deeper reasoning, longer execution and repeated verification.
These are not simply “standard” and “fast” editions. Carbon is not a lightweight model like Gemini Flash that trades some capability for speed and lower cost. Google says Carbon uses more tokens than Argon to reason deeply, explore alternatives and run verification loops.
Carbon and Argon compared
| Feature | Gemini 4 Argon | Gemini 4 Carbon |
|---|---|---|
| Official positioning | Efficient frontier reasoning | Most capable flagship reasoning |
| Main uses | Everyday development, complex analysis, knowledge work | Difficult coding, complex research, long-running agents |
| Reasoning | Balances speed and depth | More exploration and verification |
| Response speed | Usually faster | Complex tasks may take longer |
| Token use | Relatively lower | Google says it generally uses more tokens |
| Google AI Pro | Available | Not included in the personal plan |
| Google AI Ultra | Available | Available |
| Paid API | Available | Available |
Carbon's main advantage is tackling difficult problems, not answering ordinary questions faster. Argon is usually enough to explain Python code, organize meeting notes or summarize a report. Carbon may be more useful when a recurring bug spans a large project and requires tracing the cause across modules, then designing, implementing and verifying a fix.
These recommendations follow Google's positioning; they are not independent test results proving that Carbon wins in every category.
Announcement: Gemini 4 Carbon is now available.
For the broader plan comparison, see our Google AI subscription guide. For quotas, see Gemini usage limits.
2. Which plans include Carbon? Can Pro subscribers use it?
Gemini Help lists the following access rules.
Access in the Gemini app
This table applies to personal Google accounts. Work and school access depends on the Workspace license.
| Google AI plan | Gemini 4 Argon | Gemini 4 Carbon |
|---|---|---|
| Free | No | No |
| Google AI Plus | No | No |
| Google AI Pro | Yes | No |
| Google AI Ultra | Yes | Yes |
Google AI Pro is a subscription name; Gemini Pro is a model name. Google AI Plus includes access to the Gemini Pro model, but not Gemini 4 Argon. Google AI Pro does not include Gemini 4 Carbon either.
What should Google AI Pro users do?
If you already have Pro, you can keep using Argon. Google still positions it as the better fit for most everyday development and demanding office work, so Carbon's arrival alone is not a good reason to upgrade immediately.
Ultra is worth considering if you regularly work on:
- Architecture and cross-module debugging in large codebases.
- Long-running automated AI agents.
- Multistage research, repeated verification and difficult problem-solving.
- In-depth analysis across large collections of files and data.
Developers who only occasionally need Carbon can consider the paid API instead of upgrading an entire subscription for one task.
What does Ultra add?
Ultra users can choose Carbon in the Gemini app and still switch back to Argon. Access is not unlimited: Google's help documentation ties usage to the plan, remaining allowance and feature-specific limits. More capable models and higher thinking levels generally consume more resources.
Check your account's current usage limits rather than inferring a fixed message allowance from the plan name.
Source: Google's guide to Gemini models.
3. How does Carbon work in Antigravity?
Google Antigravity is an AI agent coding environment with code generation, tool calls, file editing and multistep workflows. Its model documentation lists:
| Antigravity account | Argon | Carbon |
|---|---|---|
| Free / Plus | No | No |
| Google AI Pro | Yes, subject to eligibility | No |
| Google AI Ultra | Yes | Yes |
| Enterprise | Yes | Yes |
Argon access for Pro has an extra condition: Google lists paid Pro accounts and trial accounts tied to eligible premium devices. Do not assume every Pro trial includes Argon.
/swarm: several agents working on one task
Alongside Carbon, Google announced a preview of /swarm in Antigravity. It can divide a complex task among specialized AI agents. One possible arrangement is:
- Carbon handles architecture and planning.
- Argon or another lighter model edits individual modules.
- A verification agent runs tests and checks changes.
- Carbon resolves conflicts, coordinates results and performs the final review.
This is an illustrative division of work, not a fixed execution sequence. It reserves the more capable model for difficult judgments instead of assigning every step to Carbon.
The Antigravity team warns that parallel agents can use quota faster than ordinary conversations, even with model switching and optimizations that reduce idle time. /swarm trades additional compute for parallel work on complex tasks; it does not necessarily save quota.
Sources: Antigravity model access and Carbon and /swarm announcement.
4. Does the Carbon API require Ultra?
You do not need an Ultra subscription to use Carbon through the paid API. Ultra is the access requirement in Google's personal subscription lineup, while the Gemini API has separate access and billing. Google says paid API users can use Carbon through developer tools such as Google AI Studio.
- Individuals using Carbon in the Gemini app must meet Ultra access requirements.
- Developers can use a paid Gemini API account without buying Ultra first.
- Businesses may also have access through products such as Gemini for business, depending on their license.
These entry points do not share one quota system.
Getting started in Google AI Studio
1. Open AI Studio. Sign in to Google AI Studio with your Google account.
2. Check API billing. Carbon has no free API tier. Your project and API key need valid paid access.
3. Select the model. Look for gemini-4-carbon in the supported model list. Argon's model ID is gemini-4-argon.
4. Run a test task. Verify access in AI Studio's model testing interface before integrating it into your application.
5. Set a budget and monitor usage. Check costs and quotas in the relevant billing and usage screens before production use. Automated agents may call a model repeatedly within one task and use far more tokens than a normal chat.
If Carbon is missing, the subscription type is not necessarily the cause. Regional rollout, project permissions, paid tier or gradual availability may be involved. Use the account screen and API response to diagnose the issue.
5. How much does the Carbon API cost?
Google has introduced promotional pricing for Carbon. Through November 30, 2026, Carbon and Argon have the same standard API rates.
Standard API prices
All amounts below are in US dollars per million tokens.
| Model and token type | Through November 30, 2026 | From December 1, 2026 |
|---|---|---|
| Carbon input | $2 | $4 |
| Carbon output | $10 | $20 |
| Argon input | $2 | $4 |
| Argon output | $10 | $20 |
Both models' standard input and output rates double on December 1, 2026, under Google's announced pricing schedule. Equal rates, however, do not guarantee equal costs for the same task.
Why might Carbon cost more?
API bills depend on token use as well as the rate. Google says Carbon may spend more tokens reasoning through and verifying difficult tasks. Output billing includes thinking tokens—the model's internal reasoning—so the visible answer's length is not enough to estimate cost.
Consider this hypothetical task:
| Item | Argon | Carbon |
|---|---|---|
| Input tokens | 100,000 | 100,000 |
| Output and thinking tokens | 20,000 | 60,000 |
| Input cost | $0.20 | $0.20 |
| Output cost | $0.20 | $0.60 |
| Total | $0.40 | $0.80 |
This example uses November's promotional rates; it is not a real model test. Carbon costs twice as much here despite identical unit prices.
Carbon could also reduce retries, manual corrections and follow-up calls by solving a problem more effectively, lowering the overall cost. For developers, the useful measure is the total cost of completing a task to the required standard, not just the price per million tokens.
Batch processing, caching and tool charges
These have separate billing rules. Carbon's promotional Batch API rates are:
- Input: $1 per million tokens.
- Output: $5 per million tokens.
Those rates apply to eligible batch tasks, not every real-time request. For implicit context caching, cached input costs $0.10 per million tokens for both models during the promotional period, rising to $0.20 on December 1, 2026. Both model documents list implicit caching support but no explicit caching support.
Applications using Google Search Grounding, Maps Grounding or other tools also need to account for the relevant additional charges.
Current rates: Google AI Studio API pricing.
6. Which model fits your work?
Google's current positioning does not suggest making Carbon the default for everything.
Everyday chat, writing and organization
Start with Flash, Gemini Pro or a model already in your plan for tasks such as:
- Summarizing a news article.
- Editing emails and copy.
- Organizing ordinary meeting notes.
- Answering general knowledge questions.
Carbon is usually unnecessary here; its extra resource use may not bring a matching benefit.
Everyday software development
Start with Argon for:
- Editing functions.
- Writing tests.
- Refactoring one module.
- Investigating common code errors.
- Understanding ordinary program logic.
Google positions Argon for most software engineering and knowledge work, making it a sensible baseline for developer tests.
Large engineering projects and difficult bugs
Consider Carbon for:
- Tracing hard-to-reproduce bugs across services.
- Migrating large codebases.
- Redesigning system architecture.
- Development tasks requiring sustained planning, execution and testing.
These tasks require more than code generation: a model must retain long-term goals, handle complex dependencies and repeatedly verify its work. Carbon's stated role is closer to those needs.
Scientific research and complex data analysis
Include Carbon in a comparison test for:
- Analysis across many papers and technical sources.
- Developing and verifying multistage research plans.
- Difficult mathematical reasoning.
- Combining charts, documents and structured data.
Deeper reasoning may help, but key conclusions still need independent checks. A more capable model does not remove the need for human review in finance, law or scientific research.
7. How should you read Google's benchmark results?
Google's announcement highlights Carbon's results in complex coding, agent workflows and scientific reasoning, including reported leading results on CursorBench 4.0, Perplexity WANDR and Zapier AutomationBench.
Benchmarks do not cover every real task. A lead on a particular test does not establish better Chinese writing, everyday answers or results in your workflow.
Evaluate accuracy alongside resource use. Carbon may achieve better results by spending more compute. That trade-off may not pay off on a straightforward task.
Launch materials are not independent hands-on reviews. Even when third parties supply or run some benchmarks, Google's selection and presentation of results need to be read alongside the testing methods.
As of this article's update, we have not run matched-task tests of Carbon and Argon. We therefore do not offer unverified speed rankings, Chinese-language scores or fixed quota-consumption ratios.
8. Should you upgrade to Ultra?
If you already have Pro, consider three questions:
Does Argon regularly fail to finish your tasks? If most work succeeds on the first attempt, Carbon alone is not an obvious reason to upgrade.
Do you often run workflows that require extended reasoning? Daily cross-module development, research analysis or complex agent automation makes Carbon more relevant.
Can you try the paid API on a small scale first? Developers can compare both models on the same real tasks before changing subscriptions:
- Pick 10 tasks you regularly perform.
- Give both models the same input.
- Apply the same acceptance criteria.
- Record completion, time spent on manual corrections and total token use.
- Compare the total cost per acceptable result.
That is more useful than a model leaderboard alone. If you do not use APIs, base the decision on how often you need deep reasoning and long-running tasks.
9. Frequently asked questions
Q1: Is Gemini 4 Carbon free?
Under the current rules, Carbon is unavailable to free Gemini app users and has no free API tier. Individuals need eligible Ultra access; developers can use the paid API.
Q2: Can Google AI Pro use Carbon?
Pro currently provides Argon, not Carbon, in the Gemini app and Antigravity. Carbon requires an upgrade to Ultra or separate paid API access.
Q3: Is Carbon faster than Argon?
Not necessarily. Google positions Argon as the more efficient frontier reasoning model, while Carbon spends more resources on difficult problems. Argon may respond faster on ordinary tasks; the actual difference needs testing.
Q4: Are their API prices the same?
Their published standard rates match: through November 30, 2026, input is $2 and output is $10 per million tokens. From December 1, these rise to $4 and $20. The total cost of a task can still differ.
Q5: Can Carbon use up quota faster?
Yes, it may. Google says Carbon uses more tokens for deep reasoning and verification. More capable models and higher thinking levels also consume more resources. Actual usage depends on the task, settings, access method and plan; there is no universal multiplier.
Q6: Can Antigravity free users access Carbon?
Not through the regular free tier. Carbon access in Antigravity is mainly for Ultra and eligible Enterprise users. Google also has an application program for eligible scientists that may offer six months of high-quota access. It requires application and selection and is not part of the standard free allowance.
Q7: When does the API promotion end?
The published promotion runs through November 30, 2026, with new prices starting December 1. Check current rates before deploying to production so that long-term estimates do not rely on temporary prices.
10. Choosing between Carbon and Argon
For most Pro subscribers, Argon remains the sensible starting point, covering everyday development and knowledge work with a balance of speed and resource use. Ultra subscribers can switch to Carbon when a difficult task warrants it.
API developers currently pay the same standard unit rates for either model. Compare total consumption, completion rates and manual correction costs in your own workflow. Pro includes Argon, while Carbon in the Gemini app requires Ultra; Carbon's higher capability does not mean every task will be faster or cheaper. Both models' standard API rates rise on December 1, 2026.
Keep Argon for the work it handles well and use Carbon where deeper reasoning and longer execution earn their cost.
This article uses Google's public documentation available as of November 11, 2026. We have not independently tested regional access for individual accounts, usage allowances, response speed or matched-task costs. If access or billing rules change, consult Google's current pages.