<p>Google has unveiled Gemini 4 Argon, its latest frontier artificial intelligence model, but most people cannot use it yet.</p><p>&nbsp;</p><p>The model is initially rolling out to a group of trusted cyber defenders through Google's Fairwind Program. Google says the restricted launch is part of a phased approach to testing the model's capabilities and strengthening safeguards before wider access. Argon is designed for long, complex tasks across software engineering, finance, legal work and cybersecurity. Google says it is already being used internally, where teams have used Argon agents to analyse data-centre memory and identify optimisations that could free up more than 300 TiB of memory once deployed.</p><h2><strong>Gemini 4 Argon gets a huge 1-million-token output limit</strong></h2><p>One of the biggest changes is the model's ability to generate up to 1 million tokens in a single output, compared with 64,000 previously. That matters because Argon is designed to work through lengthy, multi-step problems rather than simply answer individual prompts. Google says its engineers are using it for debugging, algorithm design and large-scale codebase migrations. The company also says Argon achieved a 77.9% score on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks. Google reported leading results in several other professional and business evaluations as well.</p><p>&nbsp;</p><p>The model has also been used for large software migrations, including work moving C/C++ code towards Rust. Google says one Argon-assisted project produced a Rust video decoder that ran 2.7 times faster than an existing Rust implementation while producing identical video output.</p><h2><strong>Its cybersecurity ability is the reason for the caution</strong></h2><p>Argon's most sensitive capability may be cybersecurity. Google says the model can find, validate and patch critical software vulnerabilities autonomously. It is already being used by cybersecurity company Wiz through its Scan for Good initiative. That ability has an obvious dual-use problem. The same technology that can help defenders discover vulnerabilities could potentially be misused to find weaknesses for malicious purposes.</p><p>Google therefore plans to provide Argon without some cyber guardrails only to trusted defenders and internal teams. The company says it is continuing work on safeguards against misuse, prompt injection and model misalignment before expanding availability.</p><h2><strong>When will Gemini 4 Argon be available and how much will it cost?</strong></h2><p>Google says Argon will eventually become available to developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. The introductory API price is $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95% discount during the introductory period. After that period, Google says pricing will rise to $4 per million input tokens and $20 per million output tokens. For now, however, Gemini 4 Argon's most powerful capabilities remain behind a controlled rollout. Google says feedback from its first group of cyber defenders and trusted testers will help determine the next stage of availability.</p>

Google has launched Gemini 4 Argon, a frontier AI model designed for complex coding, enterprise work and cybersecurity. It supports up to 1 million output tokens and can autonomously find and patch vulnerabilities, but access is initially limited to trusted cyber defenders.

Google has unveiled Gemini 4 Argon, its latest frontier artificial intelligence model, but most people cannot use it yet. The model is initially rolling out to a group of trusted cyber defenders through Google's Fairwind Program. Google says the restricted launch is part of a phased approach to testing the model's capabilities and strengthening safeguards before wider access. Argon is designed for long, complex tasks across software engineering, finance, legal work and cybersecurity. Google says it is already being used internally, where teams have used Argon agents to analyse data-centre memory and identify optimisations that could free up more than 300 TiB of memory once deployed. Gemini 4 Argon gets a huge 1-million-token output limit One of the biggest changes is the model's ability to generate up to 1 million tokens in a single output, compared with 64,000 previously. That matters because Argon is designed to work through lengthy, multi-step problems rather than simply answer individual prompts. Google says its engineers are using it for debugging, algorithm design and large-scale codebase migrations. The company also says Argon achieved a 77.9% score on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks. Google reported leading results in several other professional and business evaluations as well. The model has also been used for large software migrations, including work moving C/C++ code towards Rust. Google says one Argon-assisted project produced a Rust video decoder that ran 2.7 times faster than an existing Rust implementation while producing identical video output. Its cybersecurity ability is the reason for the caution Argon's most sensitive capability may be cybersecurity. Google says the model can find, validate and patch critical software vulnerabilities autonomously. It is already being used by cybersecurity company Wiz through its Scan for Good initiative. That ability has an obvious dual-use problem. The same technology that can help defenders discover vulnerabilities could potentially be misused to find weaknesses for malicious purposes. Google therefore plans to provide Argon without some cyber guardrails only to trusted defenders and internal teams. The company says it is continuing work on safeguards against misuse, prompt injection and model misalignment before expanding availability. When will Gemini 4 Argon be available and how much will it cost? Google says Argon will eventually become available to developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. The introductory API price is $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95% discount during the introductory period. After that period, Google says pricing will rise to $4 per million input tokens and $20 per million output tokens. For now, however, Gemini 4 Argon's most powerful capabilities remain behind a controlled rollout. Google says feedback from its first group of cyber defenders and trusted testers will help determine the next stage of availability.

<p>Google has launched Gemini 4 Argon, a frontier AI model designed for complex coding, enterprise work and cybersecurity. It supports up to 1 million output tokens and can autonomously find and patch vulnerabilities, but access is initially limited to trusted cyber defenders.</p>

Gemini 4 Argon: Google’s new AI can find and fix software flaws on its own

cyber-defenders

Cyber Defenders, News on Cyber Defenders, stories on Cyber Defenders, articles on Cyber Defenders, Cyber Defenders stories at Wion, Cyber Defenders news, Cyber Defenders latest news