Gemini 4 Carbon: Google's most powerful model, explained
By the ChatGBot Team · Published November 13, 2026
Google has opened up Gemini 4 Carbon, its strongest Gemini model so far. It was announced on October 28 for security teams first, and since November 10 it is available to paying developers, businesses and Google AI Ultra subscribers. Here is what Carbon is, how it compares to Gemini 4 Argon, what it costs and who should use it.
Every top AI in one app
GPT, Claude, DeepSeek, Qwen and GLM. One subscription.
Try ChatGBot →Gemini 4 Carbon at a glance
| Gemini 4 Carbon | |
|---|---|
| Made by | |
| Announced | October 28, 2026 (security teams first) |
| Generally available | November 10, 2026 |
| Model ID | gemini-4-carbon |
| Context window | 1,048,576 tokens |
| Max output | 65,536 tokens by default, more with long decode |
| Input | Text, images, video, audio, PDF |
| Output | Text |
| API price | $2 / $10 per 1M tokens until Nov 30, then $4 / $20 |
| In the Gemini app | Google AI Ultra |
What is Gemini 4 Carbon?
Carbon sits above Gemini 4 Argon, which launched in October. Google describes it as the model for the hardest jobs: large coding projects, agents that work for a long time on their own, and security work such as finding and fixing flaws in software. It can think at three levels (low, medium and high), so you can trade speed for depth.
In the Gemini app, the model picker now labels Argon "Code & analysis" and Carbon "Most complex work". That sums up the difference well.
What Carbon is good at
- Hard coding and agents. Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at more than 2x lower cost per task than the next best outside model.
- Security. Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 71.0% on CWE-bench for patching them and 54.8% on ExploitGym.
- Long documents and media. With a context window of about one million tokens, it can read large codebases, long PDFs, video and audio in one go.
- Tools. In the API it supports function calling, structured outputs, code execution, computer use, Google Search and Maps grounding, and file search.
Benchmarks and safety
At launch, Google said Carbon scored 59 on the Artificial Analysis Intelligence Index, the top score at the time. Google also reports clear safety gains over Argon:
| Test (reported by Google) | Gemini 4 Carbon | Gemini 4 Argon |
|---|---|---|
| Honesty tasks passed | 92% | 70% |
| Misuse breach rate per attempt | 0.8% | 3.1% |
| Prompt injection attack success (Gray Swan) | 0.0% | 0.7% |
| Automated cyberattack query success (Gray Swan) | 2.17% | 9.86% |
These numbers come from Google's own launch posts. Independent tests may differ, and the rankings change quickly as new models come out.
Gemini 4 Carbon vs Gemini 4 Argon
| Gemini 4 Carbon | Gemini 4 Argon | |
|---|---|---|
| Best for | The hardest coding, agent and security work | Most everyday coding and knowledge work |
| Speed | Slower, thinks longer | Faster |
| Token use | Uses more tokens on hard tasks | Lighter |
| Gemini app plan | AI Ultra | AI Pro and up |
Google itself says Argon "remains the ideal choice" for most software and knowledge work. Pick Carbon when a task is long, difficult or high stakes, and Argon when you want quick answers.
Price and how to get access
- Gemini API: available on the paid tier in Google AI Studio. The introductory price is $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that it is $4 and $20. Cached input costs 95% less.
- Gemini app, AI Mode in Search and Antigravity: Google AI Ultra subscribers.
- Businesses: through Gemini for business, the API, Antigravity and CodeMender.
- Security teams and scientists: Google runs a program for cyber defenders and takes applications from scientists for free six-month access.
Should you use it?
If you build software, run long agent tasks or work in security, Carbon is worth testing, especially while the introductory price lasts. For writing, research and everyday questions, a faster model will feel better and cost less. Many people use several models side by side and pick the one that fits each task. For a wider view of this month's launches, see our November 2026 AI model releases roundup, and for the Claude side of the race, read about Claude Mythos 5.2.
Use top models in one place
Ready to try it yourself?
Switch between top AI models in one chat.
Try ChatGBot →Frequently asked questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google's most capable Gemini model, announced on October 28, 2026 and generally available since November 10, 2026. It is built for hard, long-running work such as complex coding, agents and cybersecurity.
Is Gemini 4 Carbon better than Gemini 4 Argon?
On hard tasks, yes: Google reports higher scores and safer behavior than Argon. But Carbon is slower and uses more tokens, and Google still recommends Argon for most everyday coding and knowledge work.
How much does Gemini 4 Carbon cost?
In the Gemini API it costs $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that the price is $4 input and $20 output per million tokens. Cached input is 95% cheaper.
Can I use Gemini 4 Carbon in the Gemini app?
Yes, if you have a Google AI Ultra subscription. AI Pro subscribers keep Gemini 4 Argon.
What is the Gemini 4 Carbon context window?
Up to 1,048,576 tokens. Output is 65,536 tokens by default and can go higher with long decode. It reads text, images, video, audio and PDFs and answers in text.
Is Gemini 4 Carbon available in ChatGBot?
No. ChatGBot offers GPT, Claude, DeepSeek, Qwen and GLM models in one app. For hard coding and analysis you can use models such as Claude and GPT there.