New Gemini Models Explained: Gemini 4 Hydrogen vs Argon vs Carbon

The new Gemini models, in short: Google’s current lineup is Gemini 4, and it has three models. Gemini 4 Hydrogen is the fast everyday model, Gemini 4 Argon is the frontier model for most advanced work, and Gemini 4 Carbon is the strongest model for the longest and hardest tasks. All three shipped between September 30 and November 12, 2026.

This guide explains what each model is for, when it launched, what it costs in the Gemini API, which Google AI plan unlocks it in the Gemini app, and how the three compare on the benchmarks Google has published. Everything here was checked on November 13, 2026.

Disclosure: this article was written with an AI assistant from Google’s launch posts, developer documentation and help pages, which are linked throughout and listed under Sources. Benchmark figures are Google’s own unless stated otherwise. We have not run independent tests on these models yet.

Table of contents

• The Gemini 4 lineup at a glance
• Release timeline: July to November 2026
• Gemini 4 Hydrogen
• Gemini 4 Argon
• Gemini 4 Carbon
• API pricing compared
• Which Gemini plan gets which model
• Benchmarks
• What developers need to change
• Which new Gemini model should you use?
• Frequently asked questions

The Gemini 4 lineup at a glance

Gemini 4 uses element names in place of the familiar Pro and Flash labels. Hydrogen takes over the job that Flash models used to do, while Argon and Carbon sit at the top of the range.

Release timeline: July to November 2026

Google released new Gemini models almost every month in the second half of 2026. The Flash line moved first, and Gemini 4 followed.

• July 21: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and the security-tuned Gemini 3.5 Flash Cyber. At that point Gemini 3.5 Pro, which Google had promised for June, was still in testing, as PCWorld reported.
• August: Gemini 3.7 Flash.
• September 2: Gemini 3.8 Flash and 3.8 Flash Cyber.
• September 30: Gemini 4 Argon announced. Access started with a group of trusted cyber defenders in Google’s Fairwind Program.
• October 13: Argon opened to paying developers and to Google AI Pro and Ultra subscribers.
• October 28: Gemini 4 Carbon announced, again with cyber defenders first.
• November 10: Carbon opened to paying developers, businesses and Google AI Ultra subscribers.
• November 12: Gemini 4 Hydrogen launched for everyone, including free tiers.

Google lists every one of these launches on its Gemini models page.

Gemini 4 Hydrogen: the new everyday model

Hydrogen is the model most people will actually use. In its launch post, Google positions it for quick execution in everyday coding and agent tasks, and as the model that carries out the steps when Carbon or Argon does the planning.

It succeeds Gemini 3.8 Flash, and Google’s comparison is against that model:

• Speed: tasks finish about 2.5 times faster at the default high thinking level. Partner tests showed 2 to 2.8 times faster runs with 12% to 23% fewer tokens.
• Accuracy: fabricated claims fall by more than five times.
• Reasoning: Hydrogen scores 51 on the Artificial Analysis Intelligence Index at high thinking, one point above Argon’s 50, and 54 at the new xhigh level.
• Thinking levels: four of them (low, medium, high and xhigh), with high as the default.

Where you can use it: the Gemini API and Google AI Studio on every tier including the free one, the Gemini app, Google Search for all users, and Google Antigravity for AI Pro and Ultra subscribers. Gemini in Google Sheets gets it too, starting with Pro and Ultra.

One caution on price. Hydrogen’s $1.50 and $7.50 rates are not lower than what Gemini 3.8 Flash costs today, because Google Cloud’s pricing page still lists 3.8 Flash at an introductory $0.75 and $3.75 through December 31, 2026. Whether Hydrogen saves you money depends on how many fewer tokens it needs for your tasks.

Gemini 4 Argon: the frontier model for most advanced work

Argon was the first Gemini 4 model. Google built it for long, multi-step work in software engineering, finance, legal, cybersecurity and science, and it is the model Google recommends for the majority of software engineering and knowledge work.

• Long outputs: Argon introduced the unified 1M-token limit. Earlier Gemini models stopped at 64K output tokens.
• Knowledge work: Google’s October 13 post says Argon ranked first on the Vals Index, which covers finance, coding, legal and tax tasks.
• Inside Google: teams using Argon in Antigravity submitted over 35% more agent-written code and rolled back nearly a third less of it, according to the same post.
• Safety: Google reports shortcut attempts down to 2% on coding benchmarks and a 0.7% attack success rate on the Gray Swan indirect prompt injection benchmark.

Argon does not lead everywhere. In Google’s own launch table, as reproduced by Lindy, it scored 77.9% on DeepSWE v1.1, ahead of its rivals, but trailed Claude Opus 5.5 on Terminal-bench 4.0 (57.4% against 66.4%) and GPT-6 Astra on FrontierSWE v2 (55.0% against 65.5%).

Where you can use it: the paid tier of the Gemini API, the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers, and Google Antigravity for Ultra subscribers.

Gemini 4 Carbon: Google’s strongest model

Google announced Carbon two weeks after Argon went public and opened it to developers and subscribers on November 10. Google describes it as its strongest model, built for work that tests the limits of what a model can do: multi-step coding and debugging, open-ended scientific research, cybersecurity and long-running agent workflows.

• Top of the index: at its announcement Carbon ranked first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, per Google’s post.
• Cybersecurity: 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym and 71.0% on CWE-bench for patching.
• Honesty: Google says Carbon makes about half as many unsupported claims as Argon. On a 375-task agent honesty benchmark it passed 92% of tasks, against 70% for Argon.
• Cost of depth: Carbon uses more tokens than Argon so it can explore, reason and verify for longer. The same task will usually cost more on Carbon even though the per-token price is identical.

Where you can use it: the paid tier of the Gemini API and Gemini for business. In the Gemini app, AI Mode in Search and Google Antigravity it is limited to Google AI Ultra subscribers, according to the availability post.

Gemini 4 API pricing compared

Prices are in US dollars per million tokens. Output prices include thinking tokens.

Hydrogen’s rate is flat across the whole 1M-token window, so there is no surcharge for long prompts.

What a real request costs

Take a request with 200,000 input tokens and 20,000 output tokens, which is roughly a large codebase plus a long answer.

• Hydrogen: $0.30 input plus $0.15 output, or $0.45.
• Argon or Carbon until November 30: $0.40 plus $0.20, or $0.60.
• Argon or Carbon from December 1: $0.80 plus $0.40, or $1.20.

The introductory price for Argon and Carbon doubles on December 1. If you plan to evaluate either model through the API, November is the cheapest month to do it.

Which Gemini plan gets which model

In the Gemini app, model access depends on your Google AI plan. This table follows Google’s help page on models and plans for personal accounts.

A few details matter in practice:

• Free users cannot pick Hydrogen. With auto model selection on, most prompts go to Flash-Lite and prompts that need more reasoning may go to Hydrogen. With it off, every answer comes from Flash-Lite.
• Limits scale with the plan. Google’s limits page gives AI Plus twice the standard limits, AI Pro four times, and AI Ultra five or twenty times the AI Pro limits depending on the tier.
• Context also scales. The app reads up to 32K tokens without a plan, 128K on AI Plus and 1 million on AI Pro and Ultra.
• Stronger models drain limits faster. The app’s default thinking level is low. When a paid plan hits its limit, the chat continues on Flash-Lite.

One inconsistency to watch: Notebookcheck reports that Google told AI Plus subscribers they would lose the Pro model on November 13, while the help page still lists it. Check the model picker in your own account. Prices are the US rates on Google’s subscriptions page.

Gemini 4 benchmarks: Hydrogen vs Argon vs Carbon

These are the scores Google has published for the three models. Cells are blank where Google has not listed a number.

Read the table with three limits in mind. The numbers come from Google, not from independent testing. Some rows compare different versions of a benchmark, so the Terminal-Bench and Agents’ Last Exam figures are not like for like. And a higher index score for Hydrogen does not make it better than Argon at long tasks, because Google still points to Argon and Carbon for deep, long-horizon reasoning.

Hydrogen’s launch post adds a few scores of its own: 58.8% on HLE-Diamond, 52.3% on the Vals Tax Agent benchmark and 55.8% on the Vals Legal Research benchmark. On Terminal-Bench 4.1 it reaches 57%, compared with 18.2% for Gemini 3.8 Flash.

What developers need to change

Swapping the model ID is not the whole migration. Google’s Gemini 4 Hydrogen guide lists several differences from Gemini 3.8 Flash and earlier models.

1. One shared window. Input, thinking and output tokens share a single 1,048,576-token window. Gemini 3.8 Flash had separate 1M input and 64K output budgets.
2. Long outputs need background execution. A normal request returns up to 65,536 tokens. For more, run the request in the background and raise max_output_tokens.
3. The thinking default is higher. Hydrogen defaults to high. Argon and Carbon default to medium and stop at high. Test medium and high against your own evaluations before you settle.
4. Explicit caching is gone on Hydrogen. Implicit caching runs by default on prompts of at least 4,096 tokens. Remove cached_content fields.
5. Sampling settings are ignored. The server disregards temperature, top_p and top_k.
6. Declare tools properly. Use native function calling or structured outputs. Asking for tool calls as raw JSON text can return a MALFORMED_FUNCTION_CALL error.
7. Re-estimate your costs. Thinking tokens are billed as output.

Google also suggests pairing the models in one session: send planning turns to Carbon and execution turns to Hydrogen.

Which new Gemini model should you use?

• You use Gemini for free: leave auto model selection on. It is the only way your harder prompts reach Hydrogen.
• Everyday chat, writing and quick coding: Hydrogen. Google AI Plus is the cheapest plan that lets you choose it yourself.
• Complex coding, research, and finance or legal documents: Argon on Google AI Pro. This is the model Google recommends for most advanced work.
• Multi-day engineering projects, security work and scientific research: Carbon on Google AI Ultra. Use it when Argon falls short, because it spends more tokens on every task.
• Developers: start with Hydrogen at high thinking. Move a workload to Argon when quality is not good enough, and keep Carbon for planning and auditing roles. Finish your Argon and Carbon tests before the price change on December 1.

If you are choosing between assistants instead of between Gemini models, our ChatGPT vs Gemini vs Claude vs Grok comparison covers that decision.

Frequently asked questions

What is the newest Gemini model?

Gemini 4 Hydrogen, released on November 12, 2026. Gemini 4 Carbon, Google’s strongest model, became broadly available two days earlier.

Is Gemini 4 free?

Partly. Hydrogen has a free, rate-limited tier in the Gemini API and Google AI Studio, and free Gemini app users can reach it through auto model selection. Argon needs Google AI Pro or a paid API key, and Carbon needs Google AI Ultra or a paid API key.

What is the difference between Gemini 4 Argon and Carbon?

Both are frontier models at the same API price. Argon is the efficient choice for most advanced work. Carbon reasons for longer, uses more tokens, and leads on cybersecurity and long agent tasks.

Is Gemini 4 Hydrogen better than Gemini 3.8 Flash?

By Google’s figures, yes. It completes tasks about 2.5 times faster at its default setting and makes more than five times fewer fabricated claims. Its list price is higher than the introductory rate 3.8 Flash has until the end of 2026.

When do Gemini 4 Argon and Carbon get more expensive?

On December 1, 2026. The introductory $2 input and $10 output rate ends on November 30, and the price becomes $4 and $20 per million tokens.

The bottom line

Google’s new Gemini models split into a clear ladder. Hydrogen is the fast model for everyday work and the only one free users can touch. Argon handles most serious coding and knowledge work on Google AI Pro. Carbon is the top model for the hardest problems, at Ultra prices and higher token use. For API users the date to remember is December 1, when Argon and Carbon double in price.

Sources

• Introducing Gemini 4 Hydrogen, Google
• Gemini 4 Carbon is now available, Google
• Gemini 4 Carbon: our next frontier model, Google
• Gemini 4 Argon: Built to solve your hardest problems, Google
• The latest AI news we announced in October 2026, Google
• Gemini models, Google DeepMind
• What’s new in Gemini 4 Hydrogen, Google AI Studio docs
• Gemini API models, Google AI for Developers
• Gemini models and plans, Gemini Apps Help
• Gemini Apps limits and feature availability, Gemini Apps Help
• Google AI plans, Google
• Generative AI pricing, Google Cloud
• Gemini 4 Argon: who can use it, pricing and benchmarks, Lindy
• Gemini 4 Hydrogen: what free users and Google AI Plus get, Notebookcheck
• Google has new Gemini models, but the one everyone wants isn’t here, PCWorld

Last Updated on November 13, 2026 by Logan Carter