Gemini 4 Argon: Google's New AI Model Is Built for More Than Chat
What if an AI model could analyse videos, work through complex software problems, review large document collections, and handle multi-step tasks in one workflow?
That is the promise behind Gemini 4 Argon, Google DeepMind's frontier multimodal AI model, released to developers and businesses on October 13, 2026, after a limited preview that began on September 30.
Room for really long tasks
One of its standout features is what Google calls an industry-leading output limit of up to 1 million tokens, shared between your input and the model's reply. By default the API returns up to about 64K tokens in a single response, but longer outputs can use more of that window. Either way, it gives the model room to handle lengthy, complex tasks without being restricted to short answers.
What it's built for
Gemini 4 Argon is designed for demanding work, including:
- Software engineering: large codebase changes and complex engineering tasks.
- Legal and financial analysis: comparing evidence across documents, tables and charts.
- Multimodal input: it accepts text, images, video, audio and PDFs.
- AI agents: multi-step, "agentic" problems.
- Cybersecurity: helping find and patch critical software vulnerabilities.
The numbers
Google reports scores of 77.9% on DeepSWE v1.1 and 51.3% on AutomationBench, ahead of rival models in its own comparison table. These are Google's figures; independent testing was still to come at launch.
What it costs
For developers and businesses, introductory API pricing is $2 per million input tokens and $10 per million output tokens until November 30, 2026. From December 1, the price rises to $4 per million input tokens and $20 per million output tokens.
Why it matters
The bigger story is not simply that AI can answer questions. It is the growing ability of AI models to handle complex work involving multiple steps, different types of information, and practical technical tasks.
As these capabilities develop, developers and businesses will have new opportunities to explore AI-powered applications, automation, and software solutions.
Gemini 4 Argon could be another important step toward AI that does more than generate text.
Read Google's announcement: Introducing Gemini 4 Argon.
Source: There's An AI For That ↗