Google Releases Advanced Gemini 4 Argon AI Model with Enhanced Capabilities
Translated & summarized from Now 14 by baba
Google has released its advanced AI model, Gemini 4 Argon, featuring a million-token context window and enhanced real-world problem-solving abilities. The model demonstrated superior performance in coding tests and successfully guided a robot to play cornhole after analyzing a video. It also helped Google clean millions of lines of code in Chrome and identified a critical security flaw in hospital software in collaboration with Israeli firm Wiz. Access is available to developers and premium subscribers, with a phased rollout underway.
The story in 6 lines · by baba
- Google launched its flagship Gemini 4 Argon AI model with a million-token context window.
- The model outperformed competitors in autonomous software engineering and demonstrated advanced scientific capabilities.
- Gemini 4 Argon successfully taught a robot to play cornhole by analyzing video and coding simulations.
- It helped Google clean millions of lines of outdated code in the Chrome browser.
- The AI identified a critical security vulnerability in global hospital software with Israeli firm Wiz.
- Access is limited to developers and premium Google AI Pro and Ultra subscribers.
Google has officially launched its flagship AI model, Gemini 4 Argon, making it available to users through the Gemini app for phones and computers, as well as its AI Studio development environment. This release marks a significant shift for Google, moving from an emphasis on rapid response times to a focus on deep thinking, precision, and autonomous task execution with real-world results. The new model boasts a context window of up to one million tokens, a substantial increase from the previous generation's 64,000 tokens, allowing for better continuity in complex tasks. In autonomous software engineering tests, Gemini 4 Argon outperformed competitors like GPT-6 Astra and Claude Opus 5.5, achieving a score of 77.9% on the DeepSWE v1.1 benchmark.
Google highlighted Gemini 4 Argon's advanced scientific and problem-solving capabilities. It successfully solved a historical mathematical problem, translated a quantum error decoder to a hardware kernel that doubled calculation speed, and designed biological enzymes at record speed. In a notable demonstration, the AI analyzed a video of a cornhole game, mapped the physics, wrote simulation code, and then instructed a physical humanoid robot to successfully make seven consecutive throws.
Internally, Google utilized Gemini 4 Argon to streamline its own operations. Engineers employed the model to scan Chrome's source code, identifying and cleaning approximately six million lines of outdated code, a task expected to prevent future system crashes. The model also optimized an internal Python development environment, significantly reducing loading times and improving the performance of open-source models like Gemma.
For end-users, Gemini 4 Argon offers enhanced output generation, producing complete web applications, 3D printable files, and designed presentations, moving beyond simple text summaries. The model is integrated with over 40 leading services and applications, enabling it to perform tasks like ordering groceries from a recipe or booking restaurant reservations automatically. In cybersecurity, Gemini 4 Argon collaborated with the Israeli company Wiz to identify a critical security vulnerability in medical software used globally, a flaw missed by other models. This vulnerability exposed a master key, granting full administrative access to hospital networks.
Access to Gemini 4 Argon is now open to developers via the Gemini API and the Antigravity environment, with introductory pricing for API usage. For consumers, access is limited to subscribers of the premium Google AI Pro and Ultra packages, excluding users of the AI Plus plan. The rollout is phased, meaning even paying users may experience a delay before the Argon model becomes available in their accounts.