Google launched its new Gemini 4 Argon model globally on October 13, 2026, re-entering the race for advanced AI models. The new model focuses on executing long, multi-step tasks in software development, cybersecurity, scientific research, finance, law, and robotics, rather than just providing short chatbot answers.
Designed for Deep Work
Google describes Gemini 4 Argon as a model built for deep work on complex real-world problems. It combines advanced reasoning, visual understanding, and the ability to follow long instructions across interconnected steps. Through the Gemini API, the model supports a unified limit of one million tokens for input and output, allowing it to handle massive codebases and lengthy legal and financial documents.
Google initially provided the model to a limited group of cyber defense specialists under the Fairwind program in late September, before expanding access to developers, enterprises, and paid subscribers.
Direct Answers and Improved User Experience
According to the company, the model provides more direct answers, reducing unnecessary hedging, repetition, and common AI boilerplate. It adheres better to style guidelines and gives users clearer progress updates during long tasks.
These improvements are available in the Gemini app, where Pro and Ultra subscribers can select the model from the menu. On macOS, Google integrated the Google Antigravity interface into the Gemini app to ease the transition between chat and app building, though full access to the Antigravity platform requires an Ultra subscription.
Advanced Coding Capabilities
Agentic coding is a key strength of Gemini 4 Argon. It is designed for long software engineering tasks like code migration, large repository analysis, bug detection, and fixing security vulnerabilities without losing project context. Google claims the model scored 77.9% on the DeepSWE v1.1 benchmark, which measures performance in real-world software engineering tasks.
According to a report by The Verge, Google uses the model internally to reduce the compute power needed for the YouTube comments system and to help debug a physics simulator for Wing drones, reflecting a shift toward executing practical work within systems.
Science and Robotics Applications
The model's use cases extend to scientific research, including quantum computing error correction, nuclear fusion research, and enzyme design. Google stated that the model scored 87% on the Embodied Reasoning (ERQA) benchmark, which tests spatial and physical reasoning for applications like robot motion planning and translating video information into simulations.
The company also announced grants for 10,000 scientists and researchers, offering six months of free access to the highest tier of Google Antigravity. Applicants will be accepted gradually based on eligibility and capacity.
Executing Tasks Across External Apps
Gemini 4 Argon can run background tasks through connections with over 40 external apps and services, including Shopify, Instacart, and OpenTable. Users can switch between chat and task modes, allowing the model to transition from explaining what to do to attempting to execute the required actions semi-autonomously.
Availability and Access Channels
The model is not available for free accounts at launch, and Google has not announced a date for its arrival on lower tiers. Access requires paid subscriptions or API usage according to the following table:
| Access Channel | Availability at Launch |
|---|---|
| Gemini App | Google AI Pro and Ultra subscribers |
| AI Mode in Google Search | Google AI Pro and Ultra subscribers |
| Gemini API | Paid tier for developers and enterprises |
| Google Antigravity | Google AI Ultra subscribers and eligible enterprises |
| AI Studio and Android Studio | Via paid API tier, with extra options for Ultra subscribers |
| Free Accounts | Not available at launch |
API Pricing
Google introduced introductory pricing for the Gemini 4 Argon API, valid until November 30, 2026, alongside a 95% discount on cached input tokens.
| Item | Introductory Price | After November 30, 2026 |
|---|---|---|
| 1 Million Input Tokens | $2 | $4 |
| 1 Million Output Tokens | $10 | $20 |
The Verge noted that while the per-token price is lower than some competing models, long agentic tasks can consume large amounts of tokens before reaching a final result, meaning the nominal cost does not always guarantee lower operating expenses.
Security and Government Review
Google followed a phased approach to gather feedback and improve safeguards before the general release, especially since the model can independently detect vulnerabilities and suggest patches. The company participated in a voluntary US model review process, though a CNET report pointed out that the details of this government review are not public, limiting independent assessment of its thoroughness.
Position in the AI Competition
The release of Gemini 4 Argon positions Google against advanced models from OpenAI and Anthropic, with Google claiming it tops the Vals Index for professional tasks. However, these initial performance results come from Google, and independent testing is needed to verify quality, reliability, and token consumption. For developers and paid users, the model offers a massive context window, agentic coding capabilities, and a more direct interaction style, though it remains behind a paywall for now.
This article was written by AI.
