Google says Gemini 4 Carbon is its strongest model, but cyber defenders get it first
Google announced Gemini 4 Carbon on Wednesday, October 28, calling it its strongest model and pitching it at long-running coding work and cybersecurity defense. Developers, businesses and consumers can't use it yet: Google is giving it to a subset of cyber defenders first and says wider access will come as soon as possible.
What Google says Carbon can do
Carbon arrives two weeks after Gemini 4 Argon. Google says it ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, and that it gained 89 Elo points on WebDev Arena to reach fourth place with 1,768.
The company says Carbon is designed for long, multi-step engineering jobs. According to Koray Kavukcuoglu, Google's chief AI architect, internal teams have seen it run for days on tasks such as tracking down and patching subtle bugs across large codebases.
Google puts security at the center of the launch. It reports these scores on external cybersecurity benchmarks, measured without guardrails:
- SEC-bench Pro (finding vulnerabilities): 88.2 percent
- ExploitGym (checking whether a flaw can be exploited): 54.8 percent
- CWE-bench (patching): 71.0 percent
Google also says Carbon more than doubles Argon's success rate at writing secure code on its internal tests. It adds that the Chrome team found Carbon more than 10 times better than another lab's large frontier model at producing high-quality patches.
Safety claims
Google says Carbon uses the same layered safety system as Argon and was trained to behave better under pressure. Its own figures include:
- successful cheating on Terminal-Bench 4.0 cut to 0.0 percent, with cheating attempts at 0.4 percent;
- high-severity misalignment alerts in internal coding work down by almost 80 percent;
- about half as many unsupported claims as Argon;
- overreach down 26 percent and under-service down 29 percent;
- a 0.0 percent attack success rate on Gray Swan's indirect prompt injection benchmark, compared with 0.7 percent for Argon.
All of these figures are self-reported by Google.
Who gets it, and what it costs
Google says it is taking a deliberate approach and is part of the US government's voluntary pre-release model testing process. Starting this week, a subset of cyber defenders gets access. More trusted partners will follow through the Fairwind Program, which Google describes as early access to frontier models for defensive security work. It prioritizes governments, critical infrastructure operators and core technology platforms.
Google hasn't given a date for developers, businesses or consumers. It says it wants to keep gathering feedback and refining guardrails first.
When Carbon launches more widely, it will start at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95 percent cheaper than regular input. After the introductory period, the price rises to $4 per million input tokens and $20 per million output tokens.
What it means for you
For now, Carbon is limited to a subset of cyber defenders. If you use Gemini or build on the Gemini API, nothing changes today. Since October 13, developers have been able to build with Gemini 4 Argon on the paid tier of the Gemini API in Google AI Studio and Android Studio.
Security teams at governments, core technology platforms and critical infrastructure operators in areas such as healthcare, telecommunications, energy and finance can apply to Google's Fairwind Program. Its page says a set of partners currently get Gemini 4 Argon, and Google says Carbon access will expand to more trusted partners gradually. Everyone else should treat the benchmark numbers as Google's claims until the model is out and others can test it. Developers planning a switch should budget for the price doubling once the introductory period ends.