Google Released Gemini 4 Argon with Million-Token Context Window and Unlimited Cybersecurity Protection

• Google has unveiled Gemini 4 Argon.
• The model comes with a 1 million-token context window and new cybersecurity capabilities.
• It has also learned to work with 800,000 lines of code.

Google has unveiled Gemini 4 Argon — a new flagship model designed for complex, multi-step tasks in programming, enterprise workflows, and cybersecurity. The model is already being tested by a limited group of cyber defenders as part of the Fairwind Program, and ahead of a broader rollout, Google continues to validate its safety mechanisms. 

One of Argon’s key features is an expanded context limit of 1 million tokens — up from 64,000 in the previous version.

At launch, Gemini 4 Argon will cost $2 per 1 million input tokens and $10 per 1 million output tokens. Google plans to initially provide access to paid API customers and Google AI Ultra subscribers.

Gemini 4 Argon Helps Google in Development

Google is already using the model in internal processes, including programming, research, and working with large codebases. According to the company, Argon helped:

• Optimize a quantum algorithm, beating the published baseline result by 40%
• Free up more than 300 TiB of memory after analyzing data center telemetry, with the total estimated potential savings at 500 TiB — 1 PiB
• Scale the migration of C/C++ codebases to Rust, including the Fuchsia Zircon kernel with more than 800,000 lines

Separately, Google cited an example involving the open-source libgav1 project. Argon agents replaced 32,000 lines of SIMD code and produced a Rust version that, according to the company, runs 2.7 times faster than the previous Rust port and delivers identical video decoding output.

The model also posted strong results in specialized benchmarks. In particular, on DeepSWE v1.1, which evaluates performance on real, long-running programming tasks, Argon scored 77.9%. On AutomationBench, which tests end-to-end execution of business tasks, the model scored 51.3%.

At the same time, Google notes that Argon is not yet available to a broad user base. The company will gradually expand access while collecting feedback from testers and strengthening its safety mechanisms.

Gemini 4 Argon Strengthens Cyber Defense

Another use case for the model is cybersecurity. Google said Argon can autonomously find, validate, and patch critical software vulnerabilities.

As part of the Scan for Good initiative, Wiz is already using the model to protect critical infrastructure. During one of the tests, Argon discovered a critical vulnerability in medical software that could have exposed patients’ personal data in hospitals worldwide.

On CWE-bench v1, which evaluates models’ ability to remediate vulnerabilities, Argon scored 68% and tied for first place with another model.

Ahead of a large-scale rollout, Google is also strengthening safeguards against abuse, prompt injection attacks, and issues with aligning the model’s actions with user intent. To do so, the company uses automated and manual red teaming tests, monitoring of the model’s internal processes, and isolated environments for high-risk tests.

Recall that last week Anthropic unveiled a cheaper and faster AI model, Claude Opus 5.5, while OpenAI unveiled GPT-6 Sol and Luna with lower usage costs. 

Claude Opus 5.5 vs. GPT-6 Sol: We Gave Two New AIs One Task — and here’s What Came of It

28.09.2026

Read now

Сообщение Google Released Gemini 4 Argon with Million-Token Context Window and Unlimited Cybersecurity Protection появились сначала на INCRYPTED.