What Carbon is for
Carbon is Google's second Gemini 4 model, following Gemini 4 Argon, which reached developers two weeks earlier. DeepMind's Koray Kavukcuoglu describes Carbon as the company's strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. Google says its own teams have run Carbon for days on single tasks, such as tracking down and patching subtle bugs across large codebases and low-level systems.
The benchmark story, in context
- Artificial Analysis Intelligence Index v5.0: Carbon ranks first at 59, ahead of Anthropic's Claude Opus 5.5 at 58, OpenAI's GPT-6 Astra at 57 and Argon at 50
- SEC-bench Pro (finding real vulnerabilities): 88.2%
- CWE-bench (patching the root cause): 71.0%
- ExploitGym (checking whether a flaw is exploitable): 54.8%
Google also reports more than double Argon's success rate at generating secure code, an almost 80% cut in high-severity misalignment alerts in internal coding work, roughly half as many unsupported claims, and a 0.0% attack-success rate on Gray Swan's indirect prompt-injection benchmark. These are Google's own evaluations, run without safety guardrails on the security benchmarks, and independent reproduction has not yet been published.
Who gets it, and when
Access starts with a subset of cyber defenders, then expands to additional trusted partners through Google's Fairwind Program. Google says it is also taking part in the United States government's voluntary pre-release model testing process, and it plans broader developer, business and consumer availability as soon as possible, with no date given.
| Model | Artificial Analysis Intelligence Index | Position in this story |
|---|---|---|
| Gemini 4 Carbon | 59 | First; cyber-defence first rollout. |
| Claude Opus 5.5 | 58 | Closest rival on the index. |
| GPT-6 Astra | 57 | Ahead of Carbon on FrontierSWE 2. |
| Gemini 4 Argon | 50 | Predecessor, two weeks earlier. |
Find the root cause of this authentication bypass in the attached codebase, show the exact code path an attacker would take, then write a minimal patch and a test that proves the bypass is closed without breaking existing logins.
Frequently Asked Questions
Can I use Gemini 4 Carbon today?
Almost certainly not, unless you are in the initial subset of cyber defenders. Google is rolling it out in stages through its Fairwind Program, with developer and consumer access promised later but undated.
How much does Gemini 4 Carbon cost?
Introductory pricing reported from the announcement is $2 per million input tokens and $10 per million output tokens, with cached input discounted by 95%, rising later to $4 and $20. Confirm the current rate card in Google's own pricing pages before budgeting, since the introductory window is unspecified.
Is Carbon better than Argon at everything?
No. Carbon leads most published rows, but it trails Claude Opus 5.5 on OSWorld-2.1 and GPT-6 Astra on FrontierSWE 2 in Google's own comparison table. Its clearest lead is cybersecurity and long-horizon coding.
Why is Google releasing a security model to defenders first?
Google frames it as a deliberate deployment choice: gather feedback and iterate on guardrails with trusted defenders, while participating in voluntary government pre-release testing, before wider release.
