Google launches Gemini 4 Argon first to cyber defenders, not consumers

Key Points

1. Google unveiled Gemini 4 Argon, releasing it initially to vetted cyber defenders before developers, enterprises and consumers.
2. The model is released without cyber guardrails for trusted defenders, and raises output limits to 1 million tokens.
3. A staged rollout signals Google treating frontier cyber capability as a security release, not a product launch.

The latest:

Google’s newest frontier model is going to a small group of vetted cyber defenders before anyone else can buy it. The company said Gemini 4 Argon, announced September 30, is rolling out through its Fairwind Program and is being supplied to trusted defenders without cyber guardrails so they can use its full vulnerability-hunting capability. Broad access to developers, enterprises and consumers follows later.

Details:

• The access path: Google said Argon reaches an initial cohort of cyber defenders and trusted testers first, then expands to developers, enterprises and consumers, beginning with paid API customers and Google AI Ultra subscribers. The company said it is participating in the U.S. government’s voluntary process for pre-release model access.
• The pricing: Argon launches at an introductory $2 per million input tokens and $10 per million output tokens, according to Google, with cached input tokens discounted 95%. The company did not say how long the introductory rate holds or what the standard price will be.
• The output jump: Google said it expanded the output token limit to 1 million, up from 64,000, calling it industry-leading. The figure matters for the long-horizon tasks Google is positioning Argon around: multi-step software engineering, legal and finance knowledge work, and sustained agent runs.
• The benchmarks: Google reported state of the art on DeepSWE v1.1 at 77.9% for long-horizon software engineering and 91.7% on LVBench for long video understanding. It said Argon ranks first on Zapier’s AutomationBench with 51.3%, and leads the Vals Index measuring economic impact across finance, coding, legal and tax work.
• The security claim: Google said Argon can autonomously find, validate and patch critical software vulnerabilities, and that it ties for first on CWE-bench v1 with 68% on vulnerability remediation. Security firm Wiz is using the model through its Scan for Good initiative, which protects critical public infrastructure at no cost.
• The Wiz demonstration: In an early demonstration, Google said the model uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, a flaw it said earlier frontier models had missed. Google did not name the affected software or say whether it has been patched.
• Inside Google: The company said thousands of employees already use Argon internally. Agents analyzed fleet-wide profiling telemetry and applied memory optimizations across Google data centers, freeing more than 300 TiB with estimated total savings of 500 TiB to 1 PiB. Researchers also used it to optimize quantum computing subroutines, beating a published baseline by 40%.
• The Rust migrations: Google said Argon agents are converting C/C++ codebases to Rust internally, scaling to more than 800,000 lines for the Fuchsia Zircon kernel. For libgav1, Google’s open-source video decoder, agents replaced 32,000 lines of SIMD code and produced a memory-safe decoder running 2.7 times faster than the previous Rust port.
• The safeguards: Google listed four areas it is strengthening before broad release: misuse defense covering cyber and CBRN risk under its Frontier Safety Framework, resistance to indirect prompt injection, misalignment monitoring that watches chain-of-thought and halts execution when needed, and sealed sandboxes isolated before high-risk training.
• Who signed it: The announcement was written by Koray Kavukcuoglu, senior vice president at Google DeepMind and the company’s chief AI architect. Google framed the staged release as what safely shipping frontier capability at this level requires, rather than a limited preview tied to capacity.

Between the lines:

Two decisions sit oddly together: Google says it is strengthening four separate safety layers before broad release, while stripping cyber guardrails for trusted defenders. That is a judgment that the same capability making Argon dangerous in the wrong hands is what defenders most need now. The pricing and subscriber tiers announced alongside it suggest the commercial launch is already scheduled behind the security one.

What’s next

Watch for the opening of paid API access and Google AI Ultra availability, disclosure of the healthcare vulnerability Wiz identified, and whether the U.S. government pre-release review produces any published findings on Argon