Gemini 4 Carbon: Price, Performance and Access Status
The price, benchmark results and access status of Gemini 4 Carbon, Google's most capable model. A cost comparison with rivals and takeaways for businesses.
Contents 10
Gemini 4 Carbon is Google's most capable model by its own classification, announced on October 28, 2026. It stands out for its cybersecurity and long-horizon task abilities. For now, though, access is limited to select cyber defenders; no date has been announced for developers and businesses.
In short:
- Google announced Gemini 4 Carbon on October 28, 2026; the model stands out in cybersecurity and long-horizon tasks.
- It ranks first on the Artificial Analysis Intelligence Index v5 with a score of 59.
- The launch price is $2 per million input tokens and $10 per million output tokens; it is stated to rise to $4 / $20 afterwards.
- Access is initially limited to select cyber defenders and partners in the Fairwind Program; there is no date for general availability.
What is Gemini 4 Carbon?
Gemini 4 Carbon is the top model in Google's Gemini 4 family. Google describes it as "our most capable model, with state-of-the-art abilities in cybersecurity and long-horizon tasks." The previous model in the family, Gemini 4 Argon, became publicly available on October 13, 2026. So Carbon arrived about two weeks after Argon. You can find the difference between the two models in our Gemini 4 Argon post.
According to the notes, Carbon has a context window of 1 million tokens. That is large enough to feed a very big codebase or a long set of documents to the model in one go.
Can I use Carbon right now?
Most likely not. Google has limited access, at first, to what it calls "select cyber defenders" and to partners in the Fairwind Program. No date has been announced yet for developers, businesses and consumers. Carbon also takes part in the US government's voluntary pre-release model testing process.
The practical upshot: don't build a product on Carbon today. If you are planning ahead, use the pricing and performance data as a reference, but wait for an access date.
Performance: is Carbon really number one?
It depends. On the Artificial Analysis Intelligence Index v5, Carbon is in first place with a score of 59. Google highlights this result in its announcement post. Other independent measurements paint a more balanced picture:
- Vals Index: Carbon is second at 70.36%; Claude Opus 5.5 is right behind at 70.18%. The gap is very small.
- Coding Agent Index: Carbon is second with a score of 59. Claude Code (Opus 5.5) leads with 62. Gemini 4 Argon scores 53.
In other words, the model that is first on the general intelligence index falls behind Claude Code on the coding agent measurement. Which one is "the best" depends on what you will use it for. You can see how rival models fare on the coding side in our post on Claude Opus 5.5 in Antigravity.
What do the cybersecurity results say?
This is where Carbon really stands out. The results Google published itself cover three areas: vulnerability discovery, exploitation and patching.
| Benchmark | What it measures | Carbon |
|---|---|---|
| SEC-bench Pro | Vulnerability discovery | 88.2% |
| ExploitGym | Exploitation | 54.8% |
| CWE-bench | Patching | 71.0% |
These numbers come from Google's own announcement; there is no independent verification in our notes. Still, the distribution is interesting: the model is very strong at finding vulnerabilities and noticeably weaker at exploiting them. For the defensive side, that can be considered a good balance. It may also explain why it was opened to defenders first.
We covered real examples of AI finding security vulnerabilities in our PHP vulnerability post and our Claude Security post.
How much does Gemini 4 Carbon cost?
There is an inconsistency between sources on pricing, so it needs to be read carefully.
According to Google's announcement, the launch price is $2 per million input tokens and $10 per million output tokens. Cached input gets a 95% discount. After the launch period, the price is stated to become $4 / $20. Vals, on the other hand, lists Carbon directly at $4 / $20. The reason for the difference is probably that Vals shows the standard post-launch price. When planning, it is safer to take $4 / $20 as the baseline.
How does it compare with rivals on cost?
Artificial Analysis runs models on the same set of tasks and calculates the cost per task. The results are as follows:
| Gemini 4 Carbon | 7,76 |
|---|---|
| Claude Opus 5.5 (max) | 10,04 |
| GPT-6 Astra (max) | 6,25 |
Kaynak: Artificial Analysis
Carbon is about 23% cheaper than Claude Opus 5.5 and about 1.2 times more expensive than GPT-6 Astra. Part of the difference comes from token usage: Carbon uses 123 thousand output tokens per task, while Opus 5.5 uses 173 thousand. In other words, it finishes the same job with less "talking," which lowers the bill.
An important detail: this comparison was made at Carbon's standard $4 / $20 price. For as long as the launch price applies, the cost would be even lower.
What does this mean for businesses?
There are three practical takeaways:
- Don't rush. There is no general availability date. Rather than tying your architecture to a single model now, build a layer that lets you swap models easily.
- Measure cost per task. Token price alone can be misleading. As the Carbon example shows, a model that spends fewer tokens can end up cheaper even if its unit price is higher.
- If you have a security team, keep a close eye on it. The strong results in vulnerability discovery and patching are relevant to code review processes. A pilot trial may make sense once access opens up.
Before moving to a new model, running a small evaluation on your own workload is always the safest path. A benchmark ranking won't mirror your use case exactly.
Frequently asked questions
When will Gemini 4 Carbon be available to everyone?
Google has not announced a date. Access is first for select cyber defenders and Fairwind Program partners. No timing has been given for developers, businesses and consumers.
Is Gemini 4 Carbon better than Claude Opus 5.5?
It varies by measurement. Carbon is first on the Artificial Analysis Intelligence Index. It edges ahead on the Vals Index by a very small margin (70.36% vs. 70.18%), but on the Coding Agent Index Claude Code (Opus 5.5) leads with 62 against 59.
How much does Gemini 4 Carbon cost?
The launch price is $2 per million input tokens and $10 per million output tokens. Cached input gets a 95% discount. After the launch period, it is stated to become $4 / $20.
What is the difference between Carbon and Argon?
Argon became available to everyone on October 13, 2026 and scored 53 on the Coding Agent Index. Carbon is higher on the same index with 59; however, access to Carbon is limited for now.
Conclusion
Gemini 4 Carbon is an ambitious model, especially in cybersecurity and long-horizon tasks, and it comes in below Claude Opus 5.5 on cost per task. On the other hand, there is no date for general availability, and it trails its rival on the coding agent measurement. So the right move today is to build a flexible, model-independent setup and wait for access to open up. If you are planning how to integrate AI models into your products or business processes, the EngerekTech team would be happy to help.
Sources
