Google rolls out Gemini 4 Argon to cyber defenders before wider release

Google DeepMind has introduced Gemini 4 Argon, a frontier AI model built for software engineering, enterprise knowledge work such as legal and financial analysis, and cybersecurity defense. The company said Argon is rolling out first to a group of trusted cyber defenders through its Fairwind Program, rather than to the public, as part of what it called a phased release.

Argon is designed to sustain reasoning across long, multi-step tasks. Google expanded the model’s output token limit — a cap on how much text it can generate in one response — to 1 million tokens, up from 64,000 in its prior model. Google said thousands of its own employees are already using Argon internally, including quantum computing researchers who used it to optimize subroutines, beating a published baseline by 40% in one case, and agents that identified memory optimizations across Google’s data centers, freeing more than 300 TiB of memory so far. Google also said Argon agents are being used to migrate large C/C++ codebases to Rust, including a 32,000-line rewrite of SIMD code in the libgav1 video decoder that Google says runs 2.7 times faster than an earlier Rust port while producing identical video output.

Google reported that Argon set a new score of 77.9% on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks, and ranked first on the Vals Index, which measures economic impact across finance, coding, legal and tax work weighted by contribution to U.S. GDP. The company also cited leading results on Zapier’s AutomationBench, for business-process execution, and on LVBench, a long-video-understanding benchmark, where Argon scored 91.7%.

Built for cybersecurity defense

Google said it trained Argon specifically to find, validate and patch software vulnerabilities, and the model tied for first place on CWE-bench v1, a vulnerability-remediation benchmark, with a score of 68%. The security firm Wiz is already using Argon through its Scan for Good initiative, and Google said the model uncovered a critical vulnerability exposing personal data in hospital software that earlier models had missed. For vetted defenders and Google’s internal teams, the company plans to release Argon without the cybersecurity guardrails it applies elsewhere, so they can use its full vulnerability-discovery capabilities.

Google said it is still strengthening safeguards before a broader release, including defenses against prompt injection attacks, monitoring of the model’s reasoning process to catch misaligned behavior, and hardened sandboxed environments for testing. The company said it is also participating in the U.S. government’s voluntary process for pre-release access to frontier models.

Google has not given a timeline for when Argon will reach developers, enterprises and consumers, saying only that paid API customers and Google AI Ultra subscribers will be first in line. What happens during the Fairwind Program testing period, and whether Google’s safety claims hold up under outside scrutiny, will shape how quickly that wider rollout proceeds.