Google announced Gemini 4 Argon, a frontier language model designed to sustain deep reasoning across complex, long-horizon workflows. The model is initially rolling out through the Fairwind Program to trusted cybersecurity defenders, with access coordinated through the U.S. government’s voluntary pre-release model access process.
Argon significantly expands output token limits to 1 million, up from the previous 64K tokens, enabling the model to work through longer, more complex problems in a single trajectory. Google is pricing Argon at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the standard input token rate.
Within Google, Argon is already powering internal workflows across thousands of employees. According to the announcement, the model has achieved notable results in quantum computing optimization, memory efficiency, and large-scale codebase migrations. In quantum algorithmic optimization, Argon beat published baselines by 40%. In one memory efficiency project, the model autonomously identified optimizations that freed 300 TiB of memory across Google’s data centers, with estimated total savings of 500 TiB to 1 PiB. Argon agents have also been migrating C/C++ codebases to Rust, scaling from tens of thousands of lines to over 800K lines for the Fuchsia Zircon kernel.
For enterprise applications, Argon sets a new state of the art on DeepSWE v1.1 (77.9%), which measures real-world software engineering tasks. It ranks #1 on AutomationBench, Zapier’s end-to-end business automation benchmark, with a score of 51.3%. The model also leads on the Vals Index, which measures economic impact across finance, coding, legal, and tax work weighted by U.S. GDP contribution.
In cybersecurity defense, Argon can autonomously find, validate, and patch critical vulnerabilities. Security firm Wiz is already using it through its Scan for Good initiative, which detected a critical vulnerability in healthcare software that previous frontier models missed. On CWE-bench v1, which evaluates vulnerability remediation, Argon ties for first place with a score of 68%.
Google is releasing Argon without cybersecurity guardrails to trusted defenders and internal teams to enable full frontier-level capabilities. Before broader rollout, the company is strengthening safeguards against misuse, prompt injection attacks, and model misalignment through red team testing and adversarial training.
Key facts
- Gemini 4 Argon supports industry-leading 1M output token limit, up from 64K
- Pricing set at $2 per million input tokens, $10 per million output tokens
- Model freed 300 TiB of memory across Google data centers through autonomous optimization
- Argon achieved 40% improvement over published baselines in quantum algorithmic optimization
- Security firm Wiz used Argon to discover a critical vulnerability in widely-used healthcare software that previous frontier models missed
- Argon scores 77.9% on DeepSWE v1.1 software engineering benchmark and ranks #1 on AutomationBench with 51.3%
- Model ties for first place on CWE-bench v1 vulnerability remediation benchmark at 68%