Articles

Gemini 4: what Argon, Carbon and Hydrogen are, what they cost and who can use them

Gemini 4 is Google’s new generation of models, and it now has three. Gemini 4 Argon has been in the paid API and in the Gemini app for Google AI Pro and Ultra since 13 October 2026. Gemini 4 Carbon, the most capable, opened to the paid API and to AI Ultra subscribers on 10 November. And Gemini 4 Hydrogen, the fast one, arrived for everyone on 12 November.

By , founder of Convexify · · updated

Cover of the article “Gemini 4: Argon, Carbon and Hydrogen”, with a portrait of Javier Sáez Hurtado

What is new in Gemini 4 (15 November 2026)

In November the Gemini 4 family became complete and open to anyone who pays. This is what has changed since the last update on 29 October, checked on Google’s pages on 15 November 2026:

  • 10 November: Carbon is available. Google’s most capable model is no longer just for cyber defenders. It is in the paid API, as gemini-4-carbon, in Google AI Studio and in Gemini for business, and for Google AI Ultra subscribers in the Gemini app, Antigravity and AI Mode in Search. The AI Plus and AI Pro plans do not include it.
  • 12 November: Hydrogen, the fast model. Google completes the family with Gemini 4 Hydrogen, cheaper and built for everyday work. It is the only one of the three with a free API tier, and it is the default model in AI Mode.
  • The launch price ends on 30 November. Argon and Carbon cost $2 per million input tokens and $10 per million output in the API until then; from 1 December, $4 and $20. The official pricing page confirms it.
  • Argon is still paid only. In the Gemini app it is only on AI Pro and Ultra; the free plan and AI Plus get Hydrogen.
  • Carbon in AI Mode, outside the EU. The AI Mode help page limits Carbon to Ultra subscribers aged 18 or over, with a personal account, in a list of countries that includes no EU member state. In the Gemini app, Google publishes no country exclusions.

What changed in October?

In October, Google went from announcing Argon to opening it to anyone who pays, and it announced Carbon:

  • 13 October: Argon is available. It is in the paid Gemini API, as gemini-4-argon, in Google AI Studio, in Antigravity and in the Gemini app for Google AI Pro and Ultra subscribers, which Google is rolling out globally. Those two plans also get it in AI Mode in Search.
  • A price with a date. The launch price, $2 per million input tokens and $10 per million output, lasts until 30 November 2026. From 1 December it goes up to $4 and $20.
  • Tasks mode in the app. A toggle in the Gemini chat switches to tasks mode, where Gemini keeps working while you are away from your computer. The app also connects to more than 40 services, such as Adobe, Shopify and OpenTable, although most of them only work in English in the US.
  • 28 October: Gemini 4 Carbon. Google calls it its most capable model, with the best results in cyber security and long tasks, at first only for cyber defenders chosen by Google.

What is Gemini 4 Argon?

Gemini 4 Argon is the first frontier AI model in Google’s Gemini 4 generation. Koray Kavukcuoglu, senior vice president at Google DeepMind and Google’s chief AI architect, announced it on 30 September 2026 as a model that can keep reasoning through long, complex tasks.

Google aims it mainly at three kinds of work: real software engineering, professional work such as legal and finance, and cyber defence. It also points to its creative writing. This is not a model for better small talk. It is a model for working on the same problem for hours.

Gemini 4 Argon
The first model in the Gemini 4 generation. Google has not announced a “Gemini 4 Pro”.
Gemini 4 Carbon
The second model in the family, announced on 28 October 2026 and opened on 10 November. It is the most capable of the three.
Gemini 4 Hydrogen
The third, from 12 November 2026: the fast, cheap model for everyday work.
Fairwind programme
Google’s limited access programme for governments and trusted security partners. It started on 2 September 2026 and has more than 650 members.
Output limit
The most the model can write in one response. Argon raises it to one million tokens, from 64,000.

That output limit is the technical change Google stresses most. With one million tokens, the model can think and write hundreds of thousands of tokens in one go, which helps when you need to move a whole codebase or write a long report in a single pass. Google did not give the size of its context window in the announcement. Since the 13 October release, VentureBeat describes a shared limit of up to one million input and output tokens in the API.

CNBC’s summary of the Gemini 4 Argon announcement (1:16). Google has not published an official video of the model.

What is Gemini 4 Argon good for?

Google says Argon is built for work that takes many steps and a lot of time. It backs this up with how the company already uses it inside Google, where thousands of employees work with it. These are Google’s own examples, not independent tests:

  • Saving memory. A team of Argon agents studied usage data from Google’s data centres and made changes that free up more than 300 TiB of memory.
  • Moving code. Argon agents are porting C and C++ code to Rust, from libraries with tens of thousands of lines up to the Fuchsia Zircon kernel, with more than 800,000.
  • Defending systems. Google says Argon can find, check and fix critical software flaws on its own. Wiz already uses it in a programme that protects public infrastructure.

For a business, the type of task is what matters. Argon is aimed at jobs where today’s agents fall short because the work is too long: reviewing a long contract, going through a set of financial documents or updating an old application. I covered the basics in my guide to OpenAI agents: an agent does not answer, it works. The longer the job, the more it matters that the model does not lose track.

What do the benchmarks say? Gemini 4 Argon vs GPT-6 Astra and Claude

In the table Google published, Argon comes first or ties in 14 of 19 tests against OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 and Claude Opus 5.5. The biggest gaps are in professional work. On AutomationBench it scores 51.3%, against 42.5% for Opus 5.5 and 41.4% for GPT-6 Astra.

Read that table with care, for three reasons. First, Google made it. Second, its methodology notes that many rival scores are the ones each company reports, and Argon was tested at its highest reasoning setting. Third, some figures do not quite match the other companies’ numbers: Google gives GPT-6 Astra 58.2% on Terminal-bench 4.0, while OpenAI reported 57.9%.

Gemini 4 Argon against its rivals

Selection from the table Google published on 30 September 2026

TestGemini 4 ArgonGPT-6 AstraClaude Opus 5.5
Professional work (AutomationBench)51.3%41.4%42.5%
Finance agent (Vals Finance Agent v2)65.4%53.5%58.6%
Legal agent (Harvey)19.6%5.4%3.8%
Agentic coding (DeepSWE v1.1)77.9%74.1%74.2%
Terminal coding (Terminal-bench 4.0)57.4%58.2%66.4%
Computer use (OSWorld-2.0, partial)69.2%72.6%No score
Long video (LVBench)91.7%87.5%83.7%
Cybersecurity (CWE-bench v1)68.0%68.0%67.0%
My own selection from Google’s table and methodology. Scores reported by the companies, not checked independently.

Where Argon does not win is just as useful. In terminal coding and in PostTrainBench, a machine learning engineering test, it trails Claude Opus 5.5, which scores 66.4% and 49.3%. In computer use, FrontierSWE v2 and Terminal-Bench Science it trails GPT-6 Astra. No model wins everything. It depends on the job.

Comparison table of Gemini 4 Argon and GPT-6 Astra: professional work, agentic coding, terminal coding, computer use and API price
Gemini 4 Argon vs GPT-6 Astra in five figures. Infographic by Javier Sáez Hurtado, CC BY 4.0.

For background on the OpenAI model, see GPT-6 Astra: what it is and who can use it.

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google’s most capable model. Koray Kavukcuoglu announced it on 28 October 2026, two weeks after Argon opened up, as a model with the best results in cyber security and long-horizon work. Google says it has run it for days at a time on engineering tasks, although it does not publish a duration test.

These are its main figures:

  • Cyber security. According to Google, it finds 88.2% of the flaws in SEC-bench Pro, scores 54.8% on ExploitGym, an exploitation test (Argon: 42.0%), and 71.0% on CWE-bench, for patching.
  • Coding. On Terminal-Bench 4.0 it goes from just over 57% with Argon to 65.2%, close to the 66.4% Google gave Claude Opus 5.5 in September. The Chrome team says it produces high-quality patches more than ten times better than another lab’s large frontier model, without saying which one.
  • Hidden instructions. In Gray Swan’s indirect prompt injection test, no attack succeeded, against 0.7% with Argon.
  • Overall index. In the Artificial Analysis Intelligence Index v5.0, an independent evaluation, Carbon comes first with 59 points, ahead of Claude Opus 5.5 (58), GPT-6 Astra (57) and Argon (50).

Gemini 4 Carbon against Argon and its rivals

Figures published on 28 October 2026

TestGemini 4 CarbonGemini 4 ArgonClaude Opus 5.5GPT-6 Astra
Artificial Analysis index v5.059505857
Terminal coding (Terminal-Bench 4.0)65.2%57.1%66.4%*58.2%*
Exploitation (ExploitGym)54.8%42.0%No scoreNo score
Software engineering (FrontierSWE 2)61.2%No score62.3%65.5%
Computer use (OSWorld-2.1, offline)81.5%No score87.2%No score
My own table from Google’s announcement and the table reported by VentureBeat. The index is by Artificial Analysis; the rest are Google’s figures. *Figures from Google’s September table, which gave Argon 57.4% on Terminal-Bench 4.0 (57.1% in the October one).

Two caveats. First, apart from the Artificial Analysis index, these are Google’s figures, and in its own comparison Carbon wins 15 of 17 tests: OpenAI leads one software engineering test and Anthropic leads one computer use test. Second, the same independent analysis gives Carbon a higher hallucination rate than Argon on its AA-Omniscience test, 11% against 4%, even though Carbon gets the best overall score on it.

Like Argon in September, Carbon started with restricted access, only for a group of cyber defenders. It lasted less than two weeks: since 10 November it has been in the paid API, Google AI Studio and Gemini for business, and for consumers in the Gemini app, Antigravity and AI Mode in Search with Google AI Ultra. The cyber variant stays with the Fairwind programme.

How much does Gemini 4 cost?

Gemini 4 Argon and Gemini 4 Carbon cost the same in the API: $2 per million input tokens and $10 per million output tokens until 30 November 2026, with cached input at $0.10. From 1 December they go up to $4, $20 and $0.20. Neither has a free API tier. Gemini 4 Hydrogen costs $1.50 and $7.50, with no scheduled increase, and does have a free tier.

I checked this on the official pricing page on 15 November. In the Gemini app you do not pay per token: Argon is included in Google AI Pro and Ultra, and Carbon only in Ultra, within each plan’s usage limits, which Google does not detail per model.

API price per million tokens

Google checked on 15 November 2026; OpenAI and Anthropic on 1 October

ModelInputOutputCached input
Gemini 4 Argon (until 30 November)$2$10$0.10
Gemini 4 Argon (from 1 December)$4$20$0.20
Gemini 4 Carbon (same prices and dates as Argon)$2, later $4$10, later $20$0.10, later $0.20
Gemini 4 Hydrogen$1.50$7.50$0.075
GPT-6.1 Sol (OpenAI)$2$10$0.10
Claude Opus 5.5 (Anthropic)$4$20$0.20 (read)
GPT-6 Astra (OpenAI)$10$50$1
My own table from Google, OpenAI and Anthropic pages. Prices can change.

The pattern is simple. Argon’s launch price matches GPT-6.1 Sol, which OpenAI announced at DevDay on 29 September, and its December price matches Claude Opus 5.5. Against GPT-6 Astra, OpenAI’s most expensive model, Argon costs a fifth until 30 November.

Who can use Gemini 4, and when?

As of 15 November 2026, each model has its own way in:

  • Gemini 4 Argon, since 13 October: in the Gemini app with Google AI Pro and Ultra, in the paid API and Google AI Studio, in Gemini Enterprise and in Google Antigravity, Google’s agentic coding environment (in Spanish), where individuals need Ultra.
  • Gemini 4 Carbon, since 10 November: in the paid API, Google AI Studio and Gemini for business, and for consumers only with Google AI Ultra, in the Gemini app, Antigravity and AI Mode in Search. AI Pro does not include it.
  • Gemini 4 Hydrogen, since 12 November: for everyone. On the free plan it appears when the app picks it, it is the default model in AI Mode and it has a free API tier.

The cyber variant of Carbon, with some of its guardrails removed, stays with the Fairwind programme.

Infographic with who can use each Gemini 4 model: Argon with AI Pro and Ultra and in the paid API since 13 October, Carbon with AI Ultra and in the paid API since 10 November and Hydrogen for everyone since 12 November
Who can use Gemini 4, and since when. Infographic by Javier Sáez Hurtado, CC BY 4.0.

Fairwind is not open to everyone. Google set it up for governments and trusted partners, and members must limit access to their internal security, incident response or penetration testing teams, with measures such as multi-factor sign-in. Google gives these members a version of Argon without the cyber guardrails of the general release, so they can use its full defensive power, and on 13 October it opened the programme to hundreds more partners.

What Google has not said is whether Argon will come to the free Gemini app.

Why is Google releasing it in stages?

Google says so in its post: releasing a model this capable safely needs a phased approach. It adds that it is taking part in the US government’s voluntary process for early model access and will keep gathering feedback from early testers while it tunes the guardrails.

The underlying reason is security. A model that can find and fix software flaws could also help someone exploit them, so Google gives it first, without limits, to the people who defend systems. For everyone else, the post lists four layers of protection:

  • Misuse: safeguards against cyber attacks and chemical, biological, radiological and nuclear threats, under its Frontier Safety Framework.
  • Prompt injection: Google says Argon is its most resistant model to harmful instructions hidden in documents or web pages, and leads the Gray Swan test.
  • Misalignment: monitoring of the model’s reasoning and actions, which can stop a task if it goes beyond what the user wanted.
  • Sandboxes: high-risk training and testing in isolated, sealed environments.

The timing helps explain it. The same week, according to The Wall Street Journal as reported by Al Jazeera, OpenAI cancelled the release of GPT-6.1 Astra because it did not meet its internal safety standards. The big labs now have models capable enough that safety sets the release pace more and more. With Carbon, Google repeated the pattern, only faster: defenders first, on 28 October, and the rest of its paying customers on 10 November.

Can you use Gemini 4 in Europe?

Yes, with one exception. Argon works in the Gemini app with Google AI Pro or Ultra and in the paid API; Carbon with AI Ultra and in the paid API; and Hydrogen for free as well. Google does not publish a per-model country list for the app, so it is worth checking in your own account.

The exception is AI Mode in Search. According to its help page, checked on 15 November 2026, Carbon is only offered there to Ultra subscribers aged 18 or over, with a personal account, in a list of countries that includes no EU member state, although it does include the UK and Switzerland. AI Mode itself works across the EU, with Hydrogen as the default.

There is small print if you work from Europe. According to the Gemini app release notes, most connected apps are only available in English in the US, and the Antigravity integration in the Mac app is in English. And in the API, since 26 October, new paid projects store the requests you send to the model by default, unless you turn it off.

If you run a business in Europe, here is what I would do now:

  • Do not switch providers because of a table. Most of the Argon and Carbon figures come from Google: test Argon on your own tasks before you decide.
  • Run your tests now. Pick three or four real tasks from your business, with data that is not sensitive, and compare models. All three models can already be tested in the API, Hydrogen for free.
  • Check where your data goes. Look at regions, data processing terms and request storage in new API projects, and check that it all fits your EU obligations.
  • Watch the launch price. If you start now with Argon or Carbon, do the numbers at the 1 December price, which is double.

What early reviews say

At first, since almost nobody could use it, reactions were about the table and the rollout: VentureBeat said Google was back in first place in many tests, but with a limited release, and TechCrunch stressed its focus on defensive security work.

There is now an independent reference. In the Artificial Analysis Intelligence Index v5.0, Argon scores 50 points, behind Claude Opus 5.5 (58) and GPT-6 Astra (57), while Carbon comes first with 59. It is a good reminder that each company’s own table is not the final word.

Sam Witteveen’s video is a calm walkthrough of what Google changed, without claiming more than the launch post shows:

Sam Witteveen on what is new in Gemini 4 Argon (10:11).

If you want to go through the numbers one by one, AICodeKing reviews Google’s table test by test:

AICodeKing’s detailed look at the Gemini 4 Argon benchmarks (14:24).

How I would use Gemini 4

What interests me most about Gemini 4 is not the table. It is the focus on long tasks: Argon’s one million output tokens and the days of continuous work Google attributes to Carbon. If that holds up outside Google’s own tests, it is good news for work that today has to be split up by hand, such as going through a large case file or updating an old application.

That said, I would test it like any new model: real tasks, data that is not sensitive and a person checking the result. A model that can work for hours in a row can also be wrong for hours in a row, and the longer the job, the more it costs to find the mistake at the end.

I would also keep the line I set out in the essay on what you would let an agent do without looking (in Spanish). I would not let any model talk to other people on my behalf or touch high-value accounts, and anything that cannot be undone would need a human to approve it. Google releasing Argon and Carbon in stages, with layers of safeguards, points the same way.

Transparency: Convexify, the firm I founded, is part of the OpenAI Select Partner programme. This article compares Gemini 4 with OpenAI and Anthropic models using the three companies’ public documentation and the linked sources.

Key takeaways

  • Gemini 4 is Google’s new generation of models: Argon, from September 2026; Carbon, the most capable, opened on 10 November; and Hydrogen, the fast one, from 12 November.
  • Argon is built for long coding jobs, professional work and cyber defence, with up to one million output tokens, and since 13 October it has been in the paid API and in the Gemini app with AI Pro and Ultra.
  • Carbon comes first in the Artificial Analysis index and is in the paid API and, for consumers, only on Google AI Ultra.
  • Argon and Carbon cost $2 and $10 per million tokens in the API until 30 November; from 1 December, $4 and $20. Hydrogen costs $1.50 and $7.50.
  • In Europe all three can be used; the exception is Carbon in AI Mode in Search, which is not offered in the EU.

The article in one image

Visual summary in six ideas of the article on Gemini 4: the three models, what Argon is for, access, price, Carbon and availability in Europe
You can reuse this infographic under a CC BY 4.0 licence: credit Javier Sáez Hurtado and link to this article.

Frequently asked questions

What is Gemini 4 Argon?

It is the first model in Google’s Gemini 4 generation, announced on 30 September 2026 and available since 13 October. Google describes it as a model for long, complex workflows: real software engineering, professional work such as legal and finance, and cyber defence.

What is Gemini 4 Carbon?

It is Google’s most capable model, announced on 28 October 2026 and available since 10 November in the paid API and for Google AI Ultra subscribers. It improves on Argon in cyber security and long coding tasks: Google says it finds 88.2% of the flaws in the SEC-bench Pro test and has worked on engineering tasks for days at a time. In the independent Artificial Analysis index it comes first, with 59 points.

When is Gemini 4 coming out?

It is already here. Gemini 4 Argon has been available since 13 October 2026 in the paid API and the Gemini app with AI Pro and Ultra; Gemini 4 Carbon since 10 November in the paid API and with AI Ultra; and Gemini 4 Hydrogen since 12 November for everyone, including the free plan when the app picks it.

Can I use Gemini 4 Carbon?

Yes, since 10 November 2026, if you pay. It is in the paid API and Google AI Studio, and for consumers only with Google AI Ultra, in the Gemini app, Antigravity and AI Mode in Search (the last one outside the EU). AI Pro does not include it. In the API it costs the same as Argon.

What is the difference between Gemini 4 Argon, Carbon and Hydrogen?

Carbon is the most capable, for the hardest and longest problems; Argon is the frontier model for most complex coding and analysis work; and Hydrogen is the fast, cheap one for everyday tasks. In the API, Argon and Carbon cost $2 and $10 per million tokens until 30 November and $4 and $20 after that; Hydrogen costs $1.50 and $7.50.

Is Gemini 4 Argon free?

No. In the Gemini app it is only on Google AI Pro and Ultra; as of 15 November 2026, the free plan and AI Plus get Gemini 4 Hydrogen, not Argon. The API has no free tier for Argon either. Hydrogen is the only Gemini 4 model with a free tier.

How much does Gemini 4 Argon cost?

Until 30 November 2026 it costs $2 per million input tokens and $10 per million output tokens, with cached input at $0.10. From 1 December it goes up to $4, $20 and $0.20. Carbon has the same prices. These are the prices on the official Gemini API pricing page, checked on 15 November 2026.

Is Gemini 4 Argon the same as Gemini 4 Pro?

Google does not use the name Gemini 4 Pro. All three models in the family are named after chemical elements: Argon, Carbon and Hydrogen. “Gemini 4 Pro” is what leaks and some media called it before Argon was announced, and 9to5Google noted at the time that Argon brought a new naming scheme.

Is Gemini 4 available in Europe?

Yes. Argon is in the Gemini app with AI Pro and Ultra and in the paid API; Carbon with AI Ultra and in the paid API; and Hydrogen for everyone. The exception is AI Mode in Search: Google does not offer Carbon there in any EU country. Connected apps are mostly in English.

Sources

  1. Google, Gemini 4 Argon: our next era of frontier intelligence (30 Sep 2026)
  2. Google DeepMind, Gemini 4 Argon evaluation methodology
  3. Google, Fairwind cyber defence programme (2 Sep 2026)
  4. Google AI for Developers, Gemini API models (checked 29 Oct 2026)
  5. Google AI for Developers, Gemini API pricing (checked 29 Oct 2026)
  6. OpenAI, Introducing GPT-6.1 Sol (29 Sep 2026)
  7. Anthropic, Claude pricing (checked 1 Oct 2026)
  8. 9to5Google, Google announces Gemini 4 Argon (30 Sep 2026)
  9. TechCrunch, Google releases Gemini 4 Argon (30 Sep 2026)
  10. VentureBeat, Google unveils Gemini 4 Argon in limited release (30 Sep 2026)
  11. Al Jazeera, OpenAI scraps release of GPT-6.1 Astra (29 Sep 2026)
  12. Google, Gemini 4 Argon is now available in the Gemini app (13 Oct 2026)
  13. Gemini Apps release notes (13 Oct 2026)
  14. Google AI for Developers, Gemini API changelog (checked 29 Oct 2026)
  15. VentureBeat, Google makes Gemini 4 Argon broadly available (13 Oct 2026)
  16. Google, Gemini 4 Carbon: our next frontier model (28 Oct 2026)
  17. VentureBeat, Google unveils Gemini 4 Carbon (28 Oct 2026)
  18. Google, Gemini 4 Carbon is now available (10 Nov 2026)
  19. Google, Gemini 4 Carbon for developers (10 Nov 2026)
  20. Google, Introducing Gemini 4 Hydrogen (12 Nov 2026)
  21. Google AI Studio, Gemini API pricing (checked 15 Nov 2026)
  22. Google AI Studio, Gemini API changelog (checked 15 Nov 2026)
  23. Google, Gemini app plans (checked 15 Nov 2026)
  24. Google Search Help, AI Mode models and countries (checked 15 Nov 2026)
  25. 9to5Google, Gemini 4 Carbon comes to the Gemini app for AI Ultra (10 Nov 2026)
  26. VentureBeat, Gemini 4 Carbon is now publicly available (10 Nov 2026)

Javier Sáez Hurtado is the founder of Convexify, an Andorra-based boutique AI firm, and Head of AI Heroes at Learning Heroes. Podcast appearances · LinkedIn · YouTube channel