Key Points
- Google has announced Gemini 4 Argon, its new frontier AI model, but is releasing it first only to trusted cyber defenders through its Fairwind Program.
- Google says Argon can produce up to 1 million output tokens, up from 64,000, and will start at $2 per million input tokens and $10 per million output tokens.
- The staged launch, tied to a US government pre-release testing process, shows Google treating the model’s cyber skills as a risk as well as a selling point.
- Paid API customers and Google AI Ultra subscribers are next in line, but Google has not given a date for broader access.
Google has a new flagship AI model, and most people can’t use it yet. The company announced Gemini 4 Argon on Wednesday, Sept. 30, calling it its “next era of frontier intelligence,” and immediately limited access to a small group of trusted cyber defenders while it tightens safety guardrails.
It is an unusual way to launch a headline model, and it says a lot about where the AI race is heading.
Who gets Gemini 4 Argon first
In a post on Google’s official blog, Google DeepMind SVP and chief AI architect Koray Kavukcuoglu said Argon is rolling out first to cyber defenders through Google’s Fairwind Program. Google said it is “actively engaged in the U.S. government’s voluntary process for pre-release model access” while it gradually expands access.
The next group in line will be developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. Google gave no date for that wider release.
“Starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible,” Tulsee Doshi, Google’s Gemini model product lead, told CNBC.
Fairwind opened on Sept. 3 with the smaller Gemini 3.8 Flash Cyber model and has since signed up more than 650 organizations, including CrowdStrike and Palo Alto Networks, SiliconANGLE reported.
What Google says Argon can do
Google is pitching Argon at long, complex professional work: software engineering, legal and financial research, and cybersecurity. The biggest technical change is length. Google says it raised the model’s output limit to 1 million tokens, up from 64,000, so the model can reason and write at much greater length in a single run.
The company also published benchmark results. These are Google’s own figures and have not been independently verified:
- DeepSWE v1.1, long-horizon software engineering: 77.9%.
- AutomationBench, Zapier’s business-process benchmark: 51.3%.
- LVBench, long video understanding: 91.7%.
- CWE-bench v1, fixing security vulnerabilities: 68%, which Google says ties for first place.
CNBC reported that Argon ties rival models from OpenAI and xAI on cybersecurity benchmarks. Google also cited internal uses, saying Argon agents freed up more than 300 TiB of memory across its data centers and are helping migrate large C and C++ codebases to Rust.
Why the cyber angle drives the rollout
Google says it trained Argon to find, validate and patch software vulnerabilities on its own. For trusted defenders and Google’s internal teams only, the company says it will release Argon without its usual cyber guardrails so they can use the model’s full defensive capability. Security firm Wiz is already using it in a program that scans critical public infrastructure for exposures.
That same capability is what makes a wide release risky. Google says it is strengthening safeguards in four areas before broad availability:
- Blocking misuse for cyberattacks or chemical, biological, radiological and nuclear threats.
- Defending against prompt injection attacks that try to hijack the model.
- Monitoring for misalignment, including watching the model’s reasoning and stopping it when needed.
- Hardening the sandboxed environments used for testing.
The Verge noted that Google is launching in a limited capacity partly to make sure the model is not misaligned.
Pricing
Google says Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period, the price rises to $4 and $20 per million tokens. Google did not say how long the introductory period lasts.
The bigger picture
The timing is pointed. Argon arrived a day after Alphabet CEO Sundar Pichai joined other tech leaders in signing a voluntary AI safety accord at the White House, CNBC noted. It also landed a day after OpenAI’s DevDay, according to The Verge, as the leading labs trade releases at a rapid clip.
Staged releases built around safety testing are becoming more common as models grow more capable at hacking and autonomous action, a concern that has surfaced in recent incidents such as OpenAI’s warning about rogue AI agents.
What happens next
For now, Argon stays with the Fairwind cohort while Google gathers feedback and finishes its safeguard work. The company says it will open access to developers, enterprises and consumers “as soon as possible.” Independent testing of Google’s benchmark claims will only be possible once that wider release arrives.
