Table of Contents
Google released three Gemini 4 models in about six weeks: Argon, Carbon and Hydrogen. The names come from the periodic table, so they don’t tell you which model is bigger, faster or cheaper. This guide compares Gemini 4 Argon vs Carbon vs Hydrogen on launch dates, features, pricing and what early users are saying, so you can pick the right model for your studies or work.
The short answer: Hydrogen is the fast, low-cost model for everyday tasks and is open to everyone. Carbon is Google’s strongest model, built for the hardest reasoning, debugging and security work. Argon, the first Gemini 4 model, sits between them for long coding and knowledge-work tasks.
Gemini 4 Argon vs Carbon vs Hydrogen at a Glance
Here is a quick view of how the three models differ before we look at each one in detail.
| Â | Gemini 4 Argon | Gemini 4 Carbon | Gemini 4 Hydrogen |
|---|---|---|---|
| First announced | 30 September 2026 | 28 October 2026 | 12 November 2026 |
| Open to developers | 13 October 2026 | 10 November 2026 | 12 November 2026 |
| Best for | Long coding and knowledge-work tasks | Hardest reasoning, debugging and security | Everyday coding, agents and chat |
| Default thinking level | Medium | Medium | High |
| API price per 1M input / output tokens | $2 / $10 until 30 Nov, then $4 / $20 | $2 / $10 until 30 Nov, then $4 / $20 | $1.50 / $7.50 |
| Gemini app access | Google AI Pro and Ultra | Google AI Ultra | All users |
Gemini 4 Argon: The First Gemini 4 Model

Argon opened the Gemini 4 generation. Google describes it as a model built to keep reasoning through long, multi-step work in software engineering, legal and finance research, and cybersecurity defence.
When Was Gemini 4 Argon Introduced?
Google announced Gemini 4 Argon on 30 September 2026, but at first it went only to a group of trusted cybersecurity defenders in its Fairwind Program while Google tested its safeguards. Wider access opened on 13 October 2026, when Argon arrived in the Gemini API and Google AI Studio for paid developers, in the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers, and in Google Antigravity for Ultra subscribers.
Key Features of Gemini 4 Argon
- Very long answers: Google raised the output limit from 64,000 tokens to 1 million tokens, so the model can think and write at length in a single response. Input and output share that 1 million-token allowance.
- Long software tasks: Argon scored 77.9% on DeepSWE v1.1, a test of real-world, long-running coding work. Inside Google, Argon in Google Antigravity raised agentic code submissions by more than 35% and cut code rollbacks by nearly a third.
- Knowledge work: It ranks first on the Vals Index, which measures performance across finance, coding, legal and tax tasks.
- Video and visual understanding: It scored 91.7% on LVBench, a long-video understanding test, and can read charts and document sets.
- Cleaner writing: Google says Argon answers more directly, hedges less, skips clichés and follows style instructions better than earlier Gemini models.
- Creative coding demos: Google showed Argon turning photos of a desk into a playable racing game, with engine audio from Gemini Omni Flash.
- Regional hosting: On Google Cloud, Argon is the only Gemini 4 model offered in single regions such as Singapore (asia-southeast1) and London (europe-west2).
What Users Say About Gemini 4 Argon
Google’s launch post shared feedback from companies in its early-access programme. Deloitte’s global AI and data lead said that on long-running tasks Argon “completed successfully 100% of the time in our evals”. Box saw a 14% accuracy gain over Gemini 3.1 Pro in enterprise knowledge work, and the finance start-up Kepler said the model matched its analyst-built reference models on more than 99% of figures. Keep in mind that Google chose which partner comments to publish.
Independent data gives a more measured picture. Figures from Artificial Analysis, compiled in eesel AI’s Carbon review, put Argon (high) at 50.4 on its intelligence index with a 4.3% made-up answer rate, meaning it often admits it doesn’t know instead of guessing. The same review suggests Argon for interactive, day-to-day coding because it responds faster per turn than Carbon.
Learn to build with AI agents: Entri’s 6-month online Agentic AI Full Stack Developer Course covers frontend, backend and databases while you build agentic workflows that plan, code, test and ship features.
Gemini 4 Carbon: Google’s Strongest Model

Carbon is the model Google points to for problems that push a model’s limits. It uses more thinking tokens than Argon to explore a problem, reason through it and check its own work.
When Was Gemini 4 Carbon Introduced?
Google announced Gemini 4 Carbon on 28 October 2026 and, as with Argon, gave it first to cyber defenders through the Fairwind Program. On 10 November 2026, Google opened Carbon to all paying developers through the Gemini API and Google AI Studio, and to Google AI Ultra subscribers in the Gemini app, Antigravity and AI Mode in Search.
Key Features of Gemini 4 Carbon
- Built for the hardest tasks: Google says Argon remains the better choice for most software and knowledge work, while Carbon suits complex multi-step debugging and open-ended scientific research.
- Top agent scores: Carbon ranks first on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, at more than twice lower cost per task than the next best model from another company, according to Google.
- Agent teams with /swarm: A preview feature in Antigravity turns one prompt into a coordinated team of agents, with Carbon handling planning and review while faster models do parallel work.
- More honest output: On Google’s 375-task agentic honesty test, Carbon passed 92% of tasks against 70% for Argon, and declined to fill in missing experimental data 98% of the time.
- Security work: A cyber version, available only through Fairwind, leads tests for finding, exploiting and patching vulnerabilities. Google’s Cloud Vulnerability Research team used it to prove two bugs were exploitable, with severity comparable to the PHP2Shell vulnerabilities.
- Same price as Argon: Carbon launched at $2 per million input tokens and $10 per million output tokens, rising to $4 / $20 on 1 December 2026.
What Users Say About Gemini 4 Carbon
Partner feedback in Google’s launch post is strong. OpenCode engineer Kit Langton said his team gave Carbon a list of requirements at the end of the day, and by morning it had built a new Rust UI library from scratch and rewritten their desktop app. He called it an incredibly brainy model. HubX reported that Carbon resolved 94% of 16 real bug fixes in a codebase with 14,000 tests, with zero regressions.
Reactions from ordinary users are more mixed. eesel AI’s review gathered comments from the Hacker News launch thread. One user who tests AI models on an unfinished paper draft said Carbon raised points that other top models, including Argon and GPT-6 Astra, had missed. Another simply called it “waaay better than Argon”. The complaints were about speed and access: Artificial Analysis measured about 67 seconds before the first word of an answer, Ultra subscribers worried their quota would run out quickly, and users on lower plans found Carbon locked in the app. A few also said the names Carbon and Argon were too easy to confuse. eesel rated Carbon 4.5 out of 5 for hard reasoning but only 2 out of 5 for tasks where a person is waiting for a reply.
Interested in AI-assisted security work? Entri’s AI-Powered Cyber Security Course in Kerala runs for 9 months online and offline, with 80+ live and recorded sessions and a 1-month guaranteed internship at NeST Digital on successful completion.
Gemini 4 Hydrogen: The Fast Everyday Model

Hydrogen is the Gemini 4 model most people will actually use. It is designed for quick, everyday work: writing and editing code, running tests, calling tools, doing research and handling documents.
When Was Gemini 4 Hydrogen Introduced?
Google introduced Gemini 4 Hydrogen on 12 November 2026, two days after Carbon’s wider release. It takes over from Gemini 3.8 Flash as Google’s fast model, and unlike Argon and Carbon, it was open to everyone from launch: all tiers of Google AI Studio and the Gemini API, the Gemini app, Google Search, and Antigravity for paid Google AI Pro and Ultra subscribers.
Key Features of Gemini 4 Hydrogen
- Smarter than its price suggests: At its default high thinking level, Hydrogen scores 51 on the Artificial Analysis Intelligence Index against 50 for Argon, and reaches 54 at its highest setting.
- Much faster than Flash: It finishes tasks 2.5 times faster than Gemini 3.8 Flash, more than triples its Terminal-Bench 4.1 score (57% against 18.2%) and cuts fabricated claims by more than five times.
- Lower price: $1.50 per million input tokens and $7.50 per million output tokens, flat across its full 1 million-token context window.
- Four thinking levels: Low, medium, high (default) and a new xhigh level, so you only wait and pay for the reasoning a task needs.
- Works with Carbon: A new Orchestrator in Antigravity switches between Carbon and Hydrogen within one session. On Terminal-Bench 4.0 it reached close to Carbon’s success rate while using 24% less quota and finishing 19% faster.
- Better everyday help: In the Gemini app it sticks more closely to your tone and format, and in Search it better understands long, multi-part questions and typos.
What Users Say About Gemini 4 Hydrogen
Hydrogen is only five days old, so independent user reviews are still thin, and most impressions come from Google’s early-access partners. Andrew Carr of Cartwheel found it “thorough and skeptical in a way I haven’t seen from past Gemini models”, saying it caught two issues in an approach his team had worked on for 10 days with rival models. Box measured about twice the speed of Gemini 3.8 Flash with 12% fewer tokens, with the biggest gains in data-heavy reports. Design tool Pixelcut ran 276 tests and said Hydrogen completed every design brief while running 1.8 times faster.
There is one practical caution from developers. A Google Developer Expert cost guide points out that Hydrogen thinks at the high level by default, while Gemini 3.8 Flash defaulted to medium, so teams switching over may see higher bills unless they set the thinking level themselves.
Gemini 4 Models Compared: Features, Speed and Access
This table puts the main technical differences side by side. Index scores are from Google and Artificial Analysis, and real results will vary with your own tasks.
| Feature | Gemini 4 Argon | Gemini 4 Carbon | Gemini 4 Hydrogen |
|---|---|---|---|
| Model ID | gemini-4-argon | gemini-4-carbon | gemini-4-hydrogen |
| Thinking levels | Low, Medium (default), High | Low, Medium (default), High | Low, Medium, High (default), xhigh |
| Token limit | 1M shared input and output | 1M shared input and output | 1M shared input and output |
| Artificial Analysis index | About 50 | 60 on high | 51 on high, 54 on xhigh |
| Speed | Faster than Carbon per turn | Slowest; often a minute before the first word | Fastest; 2.5x Gemini 3.8 Flash |
| API price from 1 Dec 2026 | $4 / $20 | $4 / $20 | $1.50 / $7.50 |
| Gemini app access | Google AI Pro and Ultra | Google AI Ultra | All users |
Want to work with models like these? Entri’s online AI-Driven Data Science Course uses hands-on exercises and real projects to teach you how to build and deploy machine learning models, with placement assistance.
Which Gemini 4 Model Should You Use?

There is no single winner. The right model depends on how hard your task is, how long you can wait and what you can spend.
- You are a student or everyday user: Choose Hydrogen. It is free to use in the Gemini app and Search, fast, and good at study guides, rewrites and research summaries.
- You are building apps or AI agents: Start with Hydrogen and set the thinking level yourself, beginning at medium. Move up only if your tests show it falls short.
- You are debugging a hard problem, auditing security or doing deep research: Choose Carbon, as long as you have access and can wait for slower, more careful answers.
- Hydrogen falls short but Carbon is too slow, or your data must stay in Singapore or London: Choose Argon.
- You run multi-step agents: Combine them. Let Carbon plan and Hydrogen carry out the work, as Google’s Orchestrator does.
The Google Developer Expert guide sums this up neatly: “Hydrogen by default, Carbon by exception, Argon by evidence.”
Final Thoughts
Argon, Carbon and Hydrogen show Google moving from one flagship model to a family where each model has a clear job. Most benchmark figures so far are reported by Google, and the Argon and Carbon prices double on 1 December 2026, so try each model on a few of your own tasks before you settle on one. For most learners, Hydrogen is the easiest place to start.
Sources
- Google: Gemini 4 Argon, our next era of frontier intelligence (30 September 2026)
- Google: Gemini 4 Argon, built to solve your hardest problems (13 October 2026)
- Google: Gemini 4 Carbon is now available (10 November 2026)
- Google: Introducing Gemini 4 Hydrogen (12 November 2026)
- eesel AI: Gemini 4 Carbon review (10 November 2026)
- DEV Community (Google Developer Experts): Hydrogen, Carbon or Argon? A cost-first guide (16 November 2026)
Frequently Asked Questions
Why are the Gemini 4 models named Argon, Carbon and Hydrogen?
Google has used chemical element names for the Gemini 4 family. The names do not tell you a model’s size, speed or price. As a Google Developer Expert guide points out, Hydrogen is the lightest element, but it is not a “lite” model in the old sense.
Is it safe to use the free tier of Gemini 4 Hydrogen for college or company projects?
On the Gemini API free tier, your prompts may be used to improve Google’s products, while paid-tier prompts are not, according to this Gemini 4 cost guide. Keep personal or confidential data out of free-tier projects, and use a paid tier or Google Cloud for company data.
What happens to Gemini 3.8 Flash now that Hydrogen is out?
Hydrogen is Google’s upgrade path from Gemini 3.8 Flash. Gemini 3.8 Flash stays cheaper for now, but its Gemini API price doubles to $1.50 / $7.50 per million tokens on 1 January 2027, the same as Hydrogen (source).
Do I need to change my code when moving from Gemini 3 to Gemini 4?
Yes. Gemini 4 ignores temperature, top_p and top_k settings, and sending thinking_budget or a MINIMAL thinking level returns an HTTP 400 error. Use the new thinking levels instead (source). If you build agents in Google Antigravity, the model settings are handled for you.
Which Gemini 4 model can read Word, Excel and PowerPoint files?
Hydrogen and Carbon accept .docx, .xlsx, .pptx and OpenDocument files directly. Argon only takes PDF and plain text files, so convert other formats first if you use Argon (source).




